Oscillation Theory For Second Order Linear, Half-Linear, Superlinear and Sublinear Dynamic Equations PDF

Oscillation Theory for Second Order
Linear, Half-Linear, Superlinear and Sublinear Dynamic

Equations
Oscillation Theory for Second

Order Linear, Half-Linear,
Superlinear and Sublinear
Dynamic Equations
by
Ravi P. Agarwal
Florida Institute ()j'Technology,
Melbourne, Florida, U.S.A.
Said R. Grace
Cairo University,
Orman, Giza, Egypt
and
Donal 0' Regan

National University
Galway, Ireland
(~f1reland,
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.
A C.LP. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-6095-2
ISBN 978-94-017-2515-6 (eBook)
DOI 10.1007/978-94-017-2515-6
Printed on acid-free paper
AII Rights Reserved

2002 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2002
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered and
executed on a computer system, for exclusive use by the purchaser of the work.
Contents
Preface
Chapter 1
1.0.
1.1.
1.2.
1.3.
1.4.
Preliminaries
Introduction
Continuability, Boundedness and Convergence to Zero
Some Useful Results from Analysis and Fixed Point Theorems
Notes and General Discussions
References
Chapter 2
2.0.
2.1.
2.2.
2.3.
2.4.
2.5.
2.6.
2.7.
2.8.
2.9.
xi
Oscillation and N onoscillation of Linear

Ordinary Differential Equations
Introduction
Sturm Comparison Theorem
Nonoscillatory Characterizations and Comparison Theorems
Nonoscillation Criteria
Oscillation Criteria
Oscillation Criteria-Integral Averaging
Oscillation Criteria-Integrable Coefficients
Forced Oscillations
References
Chapter 3
1
2
10
12
12
13
13
15
25
35
40
70
82
90
91
Oscillation and N onoscillation of

Half-Linear Differential Equations
3.0.
Introduction
93
3.1.
3.2.
3.3.
3.4.
Preliminaries
Sturm-Type Comparison Theorems
Levin-Type Comparison Theorems
Liapunov's Inequality
94
94
100
105
vi
An Oscillation Criterion for Almost-Periodic
Sturm-Liouville Equations
3.6. On Zeros of Solutions of a Singular
Half-linear Differential Equation
3.7. Nonoscillation Criteria
3.7.1. Nonoscillatory Characterization
3.7.2. Comparison Theorems
3.7.3. Sufficient Conditions for Nonoscillation of Equation (3.1.1)
3.7.4. Further Results on Nonoscillation of Equation (3.1.1)
3.8. Oscillation Criteria
3.9. Oscillation Criteria-Integral Averaging
3.9.1. Oscillation Theorems for Equation (3.1.1)
3.9.2. Interval Criteria for the Oscillation of Equation (3.1.1)
3.9.3. The Weighted-Average Oscillation Criteria
3.10. Oscillation Criteria-Integrable Coefficients
3.11. Oscillation of Damped and Forced Equations
3.11.1. Oscillation of Equation (3.11.1)
3.12. Distance of Zeros of Oscillatory Solutions
3.13. Oscillation and Nonoscillation of Half-linear
Equations with Deviating Arguments
3.13.1. Some Useful Lemmas
3.13.3. Some Extensions and More Oscillation Criteria
3.13.4. Classification of Nonoscillatory Solutions
3.13.5. A Comparison Theorem
3.13.7. Nonoscillation of Solutions of Equation (3.13.127)
3.14. Notes and General Discussions
3.15. References
3.5.
Chapter 4
4.0.
4.1.
4.2.
107
113
140
140
147
148
154
163
168
168
179
188
196
205
206
207
210
218
219
223
232
237
251
253
259
267
268
Oscillation Theory for Superlinear

Differential Equations
Introduction
Superlinear Oscillation Criteria
Further Results on Superlinear Oscillations
273
273
296
vii
4.3.
4.3.1.
4.3.2.
4.3.3.
4.4.
Existence of Nonoscillatory Solutions and Comparison Results

A Necessary Condition for Oscillation
Comparison of Nonlinear Equations of the Same Form
Linearization of Oscillation Theorems
Oscillation Criteria for Certain Nonlinear Differential Equations
4.5.
4.6.
Superlinear Forced Oscillations

Oscillation of Nonlinear Differential Equations
with Nonlinear Damping Term
4.6.1. Damping with Nonlinear Growth-Oscillation Criteria
4.6.2. Damping with Nonlinear Growth-A Nonoscillation Theorem
4.8. References
Chapter 5
5.0.
5.1.
5.2.
5.3.
5.4.
5.5.
5.6.
5.7.
358
362
362
367
371
372
Oscillation Theory for Sublinear

Introduction
Sublinear Oscillation Criteria
Further Sublinear Oscillation Criteria
Linearization of Sublinear Oscillation Theorems
Nonoscillation Theorems for Sub linear Differential Equations
Oscillation Criteria for Certain Nonlinear Differential Equations
References
Chapter 6
328
328
332
339
341
376
376
387
429
433
444
453
453
Further Results on the Oscillation of

6.0.
6.1.
Introduction
Oscillation Criteria of Olech-Opial-Wazewski Type
457
457
6.1.1.
6.1.2.
6.1.3.
6.2.
6.2.1.
6.2.2.
Some Definitions and Notation

Further Results
Oscillation Criteria for Half-linear Differential Equations
Preliminary Results
Oscillation Criteria via Liapunov's Second Method
458
459
463
466
467
469
476
6.3.
viii
6.3.1.
6.3.2.
6.4.
6.5.
Preliminaries
References
Chapter 7
7.0.
7.1.
7.1.1.
7.1.2.
7.1.3.
7.2.
7.3.
7.4.
7.5.
7.6.
7.7.
8.1.1.
8.1.2.
8.1.3.
8.1.4.
8.2.
Oscillation Results for

Differential Systems
Introduction
491
Oscillation of Nonlinear Two-dimensional Differential Systems 491
Superlinear Case
493
Sublinear Case
498
Further Oscillation Criteria
500
Oscillation Theorems for Linear Differential Systems
503
Oscillation Theorems for Differential Systems with
Functionally Commutative Matrix Coefficients
533
Comparison Theorems for Operator-Valued Linear
540
Oscillation Results for Differential Systems
with Forcing Terms
549
552
References
554
Chapter 8
8.0.
8.1.
477
479
489
490
Asymptotic Behavior of Solutions of

Certain Differential Equations
Introduction
Asymptotic Behavior of Solutions of
Nonlinear Differential Equations
Asymptotic Behavior of Nonoscillatory Solutions
Bounded Asymptotically Linear Solutions
Unbounded Asymptotically Linear Solutions
Further Extensions and Improvements
Asymptotic Behavior of Positive Solutions of
Singular Emden-Fowler-Type Equations
8.2.1. Positive Solutions of Equation (8.2.1)

8.2.2. Positive Decaying Solutions of Equation (8.2.2)
558
558
559
562
567
574
584
585
595
lX
8.2.3. Positive Solutions of Singular Emden-Fowler-Type Systems

8.4. References
Chapter 9
9.0.
9.1.
9.2.
9.3.
9.4.
9.5.
9.6.
614
622
624
Miscellaneous Topics
Introduction
Comparison and Nonoscillation Results
Nonoscillatory Solutions of Forced Differential Equations
Limit Circle Criteria and Related Properties
for Nonlinear Equations
Properties of Certain Differential Equations
References
625
625
632
638
650
654
654
Chapter 10 N onoscillation Theory for Multivalued

10.0.
10.1.
10.2.
10.3.
10.4.
10.5.
Introduction
Preliminaries
Differential Inclusions
Neutral Inclusions
References
Subject Index
656
656
657
664
668
668
670-672
Preface
This book is devoted to the qualitative theory of second order dynamic
equations. In the last 50 years the Oscillation Theory of ordinary, functional, neutral, partial, and impulsive differential equations, and their discrete versions has attracted many researchers. This has resulted in hundreds of research papers in every major mathematical journal. There are
many books which deal exclusively with oscillation of solutions of differential equations. However, most of those books appeal only to researchers
who already are familiar with the subject. Thus, in an effort to bring oscillation theory to a new and wider audience, in this book we present a
compact, thorough, and self-contained account for second order dynamic
equations. An important feature of this book is that many examples of
current interest are given to illustrate the theory.
In Chapter 1, we introduce oscillation theory of second order differential
equations. Here also we provide conditions which guarantee that all solutions of a particular class of second order nonlinear differential equations
are continuable, bounded, and converge to zero. Our final section in this
chapter states several fixed point theorems which play an important role in
establishing existence criteria of nonoscillatory solutions.
It is natural to begin our discussion with second order linear differential
equations. The interest in second order linear oscillations is due, in a large
part, to the fact that many physical systems are modelled by such equations. We begin Chapter 2 with some of the most basic results in the theory of oscillations of linear ordinary differential equations of second order.
In particular, we present Sturm and Sturm-Picone comparison theorems.
Then, we provide some necessary and sufficient conditions for nonoscillation as well as some comparison theorems of Sturm type. Also we present
sufficiency criteria for nonoscillation of solutions. Next, we establish sufficient conditions for oscillation of second order differential equations with
alternating coefficients. Integral averaging techniques and interval criteria
are two of the most important concepts in oscillation theory and both are
discussed in Chapter 2. Criteria for oscillation of linear second order differential equations with integrable coefficients are also given. We conclude
this chapter by discussing the problem of forced oscillations.
xu
In recent years the study of half-linear differential equations has become
an important area of research. This is largely due to the fact that half-linear
differential equations occur in a variety of real world problems such as in the
study of p-Laplace equations, non-Newtonian fluid theory and the turbulent flow of a polytrophic gas in a porous medium. In Chapter 3, we present
oscillation and nonoscillation criteria for second order half-linear differential equations. We begin with some preliminaries on half-linear differential
equations, and then we present Sturm and Levin type comparison theorems
and a Liapunov type inequality. Next, we provide an oscillation criterion for
almost periodic Sturm-Liouville equations. In this chapter we also present
a systematic study of the zeros of solutions of singular half-linear differential equations, nonoscillation characterizations (necessary and sufficient
conditions), comparison results and sufficiency criteria for nonoscillation.
In addition we study oscillation by employing integral and weighted averaging techniques. Here, interval criteria for the oscillation of half-linear
equations will also be provided. Next, we discuss oscillation of half-linear
equations with integrable coefficients, and damped and forced equations.
We also derive lower bounds for the distance between consecutive zeros of
an oscillatory solution. Finally in this chapter, we present a systematic
study of the oscillation and nonoscillation of half-linear equations with a
deviating argument. Here, classifications of nonoscillatory solutions, and
existence results which guarantee that the solutions have prescribed asymptotic behavior will be given.
In Chapter 4, we present a oscillation and nonoscillation theory for
second order nonlinear differential equations of superlinear type. We begin
with the oscillation of superlinear equations with sign changing coefficients.
In particular, we discuss results which involve integrals and weighted integrals of the alternating coefficients, and then provide several criteria which
use average behaviors of these integrals. More general averages such as
'weighted averages' and 'iterated averages' will also be employed. Also,
we provide sufficient conditions which guarantee the existence of nonoscillatory solutions, and then we present necessary and sufficient conditions
for oscillation of superlinear equations. Oscillation results via comparison
of nonlinear equations of the same form as well as with linear ones of the
same order will also be presented. We conclude this chapter by discussing
oscillation and nonoscillation of forced and damped superlinear differential
equations.
In Chapter 5, we present oscillation and nonoscillation criteria for all
solutions of second order nonlinear differential equations of sublinear type
with alternating coefficients. Our results will involve integrals and weighted
integrals of alternating coefficients and in some results we use integral aver-
xiii
aging techniques. We also provide some new linearized oscillation results for
second order sublinear differential equations. Next, we study the EmdenFowler equations. These equations originated from the theory of gaseous
dynamics in astrophysics around the beginning of the last century, and they
also occur in the study of gas dynamics, fluid mechanics, relativistic mechanics, nuclear physics, and in chemically reacting systems. In this chapter we present criteria for nonoscillation of sublinear Emden-Fowler type
equations. Also we compare the oscillatory behavior of certain nonlinear
equations with related half-linear differential equations. We conclude this
chapter by discussing oscillation of general nonlinear differential equations.
In Chapter 6 some special techniques are used to obtain oscillatory criteria for differential equations. Firstly, we present oscillation and nonoscillation theorems for nonlinear second order differential equations by using
the method of Olech, Opial and Wazewski. Secondly, we use a variational
inequality to study the oscillation of half-linear second order differential
equations. Finally, we apply Liapunov second method to provide criteria
for oscillation of second order nonlinear equations.
In Chapter 7, we discuss the oscillatory behavior of nonlinear twodimensional differential systems and second order vector-matrix differential
equations. We begin with the oscillation of nonlinear two-dimensional differential systems. This includes the superlinear, linear, and sub linear cases.
Then, we discuss oscillation of linear second order differential systems. Two
approaches are presented. In the first approach the system considered will
be reduced to a certain scalar Riccati inequality, so that known results from
the literature can be applied. In the second approach, we will use Riccati
and variational techniques. Next, we discuss the oscillation of nonlinear
second order differential systems with functionally commutative matrix coefficients. We show that the oscillation theory of such systems can be
effectively reduced to the study of diagonal systems of scalar second order
differential equations. Then some comparison theorems of Hille-Wintner
type for second order operator-valued linear differential equations will be
presented. Finally, we discuss oscillation of second order differential systems with a forcing term.
The study of behavioral properties of solutions of differential equations
near infinity is of immense importance in applications. In Chapter 8, we
present some recent contributions on the asymptotic behavior of solutions
of second order differential equations as well as the behavioral properties
of positive solutions of singular Emden-Fowler-type equations. It is shown
that for a large class of differential equations, not only can the existence of
nonoscillatory solutions be proved, but also an explicit asymptotic form of
XlV
the nonoscillatory solutions may be provided. Next, by imposing more restrictions on the sign of the integrable coefficient of the equation, we obtain
necessary and sufficient conditions so that the solutions have a specified
asymptotic behavior as t --; 00, i.e., solutions which behave asymptotically like a nonzero constant and also those which behave asymptotically
like ct, c 1= o. For this, various averaging techniques of the type employed
in the previous chapters to study the oscillatory behavior of such equations
will be used. Then, we study existence, uniqueness and asymptotic behavior of positive solutions of singular Emden-Fowler-type equations. The
cases when the coefficient of the equation under consideration is of constant
sign, or of an alternating sign are systematically discussed. Finally, existence and nonexistence results for positive solutions of Emden-Fowler-type
systems will be given.
In Chapter 9, some special results are presented. We extend the SturmPicone theorem, obtain nonoscillation theorems for perturbed second order
nonlinear differential equations, and present a nonlinear Picone type identity which enables us to prove some Sturm-Picone type comparison theorems for nonlinear equations. Next, we study nonoscillatory solutions of
forced differential equations of second order. Then, we present some limit
cycle criteria and discllss its related properties for nonlinear second order
differential equations. Finally, we provide some properties of solutions of
very general second order differential equations.
In our final chapter, we introduce a nonoscillation theory for second
order differential and neutral inclusions. Here, the results rely on some
fixed point theorems for multivalued maps and on a compactness criterion.
We hope this book is timely and complements existing works in the
literature on oscillation theory of second order dynamic equations. It can
be used as a textbook at the graduate level and hopefully it will stimulate
further research in this area.
It is impossible to acknowledge individually colleagues and friends to
whom we are indebted for assistance, inspiration and criticism during the
preparation of this book. We must, however, express our appreciation and
thanks to Sadhna for typing and proof reading parts of the manuscript.
Ravi P Agarwal
Said R Grace
Donal O'Regan
Chapter 1
Preliminaries
1.0. Introduction
In this introductory chapter, we shall provide conditions which guarantee that all solutions of a particular class of second order nonlinear differential equations are continuable, bounded, and converge to zero. Then, we
shall state several fixed point theorems which play an important role in our
later chapters.
In what follows, we shall use the following notation: lN = {I, 2,-}
represents the set of positive integers, lR = (-00,00), lRo = [0,00), lR+ =
(0,00) and 1R - = (-00,0) are the usual sets of real, nonnegative, positive
and negative numbers respectively. Further, for a nonnegative integer m
and I ~ JR, we denote by C (I x lRm, lR) the space of all continuous
functions mapping I x JRm into JR, and for a fixed i E lN, Ci(I X JRm , JR)
represents the space of all i times continuously differentiable functions.
Consider the second order ordinary differential equation
F(t, x(t), x'(t), xl/(t)) = 0,
(1.0.1 )
where FE C([to, 00) xlR3, lR). Bya solution of (1.0.1) we mean a function
x(t), t E [tx, 00) C [to,oo) which is twice continuously differentiable and
satisfies equation (1.0.1) on the interval [tx, 00). The number tx ::::: to::::: 0
depends on the particular solution x(t) under consideration.
A nontrivial solution x(t) of (1.0.1) is said to be oscillatory if it has
arbitrarily large zeros. Otherwise, x(t) is said to be nonoscillatory, i.e.,
x(t) is nonoscillatory if there exists a tl ::::: to such that x(t) i- 0 for
t ::::: tl. In other words, a nonoscillatory solution must be eventually positive
or negative. Equation (1.0.1) is said to be oscillatory if all its solutions are
oscillatory. In our study of oscillatory behavior of differential equations
explicit knowledge of solutions will not be needed, rather we shall provide
easily verifiable conditions on the known quantities which guarantee the
oscillation.
Chapter 1
1.1. Continuability, Boundedness

and Convergence to Zero
Consider the general second order nonlinear differential equation
(a(t)x'(t'
+ h(t,x(t),x'(t + q(t)f(x(tg(x'(t = e(t,x(t),x'(t,
(1.1.1)
where a(t), q(t) E C([to, 00), lR+), f, 9 E C(lR, lR) and h, e E C([to, 00) x
lR2 , lR), and g(y) > 0 for all y E lR. It will be convenient to write
equation (1.1.1) as the system
{
X'
y'
= att) [-a'(t)y - h(t,x,y) - q(t)f(x)g(y) + e(t,x,y)].
(1.1.2)
Let p+(t) = max{p(t) , O} and p- (t) = maxi -p~t), O} so that p(t) =

p+(t) - p-(t). Define F(x) = f; f(s)ds, G(y) = fo sds/g(s) and assume
that there exists a function r(t) E C([to, 00), lR) such that
le(t,x,y)1
yh(t, x, y)
r(t),
(1.1.3)
(1.1.4)
~ 0
and there are nonnegative constants m and n such that
Iyl
g(y) ~ m
+ nG(y).
(1.1.5)
Before we prove the continuability theorem, we state the following wellknown Gronwall's inequality.
Lemma 1.1.1.
suppose that
Let
u(t)
= [to, T)
~ c+
it
be an interval of real numbers, and
q(s)u(s)ds
for
tEl,
to
where c is a nonnegative constant, and u, q E C(I, lR+). Then,
u(t)
c exp
(1:
q(s )dS)
for
tEl.
Theorem 1.1.1.
If conditions (1.1.3) - (1.1.5) hold,
t ~ to, F(x) is bounded from below, and G(y) --700 as
all solutions of equation (1.1.1) are defined for all t ~ to.
a'(t)
Iyl
~ 0 for
--700, then
Proof. Suppose (x(t), y(t is a solution of (1.1.2) with finite escape time,
i.e., there exists T ~ to such that limt->T- [lx(t)1 + ly(t)l] = 00. Since
Preliminaries
F(x) is bounded from below, F(x) 2': -K for some constant K > O.
Define
V(x, y, t) = q(t) G(y)
+ a(t) [F(x) + KJ,
then
q'(t)
a'(t) y2
1
( y )
- q2(t) G(y) - a(t)q(t) g(y) - a(t)q(t) h(t, x, y) g(y)
V'(x, y, t)
+ a(t)q(t) e(t, x, y)
y) -
g(y)
<
[q;~~!~- G(y) + a(:~2(t) ;~)
<
[q'(t)]q2(t) G(y)
r(t)
a'(t)
a2(t) [F(x)
+ K]
r(t)
+ m a(t)q(t) + n a(t)q(t) G(y).
Integrating the above inequality from to to t and noting that r(t)/ (a(t)q(t))
is bounded on [to, TJ, we obtain for t E [to, T] that
for some constant Kl > O. Now, from Gronwall's inequality, we find

1
q(t) G(y(t))
< Kl exp
:<:;
K1exp
r(s)])
(i [[q'(S)]q(s) + n a(s) ds
t
to
(iT
to
[[q'(s)]- +nr(s)] dS)

q(s)
a(s)
< K2 <
00,
where K2 is a constant. Thus, G(y(t)) is bounded on [to, T), and

hence y(t) = x'(t) is bounded on [to, T). Finally, an integration shows
that x(t) is also bounded on [to, T), and so we have a contradiction to
the assumption that (x(t), y(t)) is a solution of (1.1.2) with finite escape
time.
Remark 1.1.1.
1. If e(t, x, y) == 0 in Theorem 1.1.1, then condition (1.1.3) can be dropped.

2. We can drop the condition on a'(t) by requiring a stronger condition
on g(y), namely, that there are positive constants M and k such that
y2
g(y) :<:; MG(y)
for
Iyl 2': k.
(1.1.6)
The proof of this result involves more details than the proof of Theorem
1.1.1, so we omit it here. We also note that (1.1.6) implies (1.1.5).
Chapter 1
Now, in what follows by the term 'solution' we shall mean a continuously

continuable solution on [to, 00 ).
In the following results, we shall present sufficient conditions which guarantee that all solutions of (1.1.1) are bounded. These results also illustrate
the interplay between a(t) and g(x') in equation (1.1.1).
Theorem 1.1.2. Suppose (1.1.4) and (1.1.5) hold with n> 0,
a'(t) 2: 0 and a(t)

and
:s:
a2,
a2
t 2: to
is a constant,
(1.1. 7)
1 () q' (t)
Ie (t,x,y )I < -;a
t q(t)'
If
F(x)
--->
as
00
Ixl
--->
(1.1.8)
00,
then all solutions of equation (1.1.1) are bounded.

Proof. Since F(x) ---> 00 as Ixl ---> 00, F(x) is bounded from below,
say, F(x) 2: -K for some constant K. Letting
V(x, y, t) =
~~!~ [F(x) + K] + G(y),
we have
V'(t)
q'(t)
:s:
[F(x)
:s:
q'(t) (q(t) [F(x)

q(t) a(t)
y )
+ K] a(t) + a(t) e(t,x,y) g(y)

+ K] + .!.~)
ng(y)
Integrating the above inequality from to to t and using (1.1.5), we find
~ + n q(t) [F(x(t)) + K] <

g(y(t))
:s:
a(t)
m+nV(to)+
t
to
+ nV(t)
q'(s) ( q(s)
ly(s)J )
q(s) na(s)[F(x(s))+K]+ g(y(s)) ds.
Now, applying Gronwall's inequality, we get
J!&2L
+ n q(t) [F(x(t)) + K]
g(y(t))
a(t)
<_ Kl exp
(1 q'(s)
dS)
q(s)
t
to
K q(t)
1 q(to) ,
where Kl > 0 is a constant. Hence, nq(t)F(x(t))/a(t):s: K1q(t)/q(to),

so F(x(t)) is bounded for t 2: to. The conclusion of the theorem now
follows from (1.1.8).
Preliminaries
Theorem 1.1.3. Suppose (1.1.4), (1.1.6) and (1.1.8) hold, a(t):::; a2,
where a2 is a constant, and
00
to
[a'(s)]ds <
a ()
s
(1.1.9)
00
and le(t, x, y)1 :::; a(t)q'(t)j(Mq(t)). Then all solutions of equation (1.1.1)
are bounded.
Proof. Condition (1.1.6) implies that there exists a constant A >

such that y2 j g(y) :::; A + MG(y) for all y. Notice also that if Iyl :::; 1,
then Iyljg(y):::; B for some constant B > 0, and if Iyl 2': 1, then
Iylj g(y) :::; y2 j g(y), so
~ < B+ y2
g(y) -
g(y)
for all
y.
By condition (1.1.8), F(x) 2': -K for some constant K> 0. If M 2': 1,

we define
V(x, y, t) =
!~!~ [F(x) + K] + G(y) + A + B.
Then since M 2': 1, we have
V'(t)
= q'(t) (q(t) [F(x) + K] + ~~)

q(t)
: :;
a(t)
M g(y)
+ [a'(t)]- (q(t) [F(x)

a(t)
a(t)
L]
g(y)
~g; (~~!~[F(X)+K]+ ~(A+B)+G(Y))

+ [a'(t)]- (q(t) [F(x)
a(t)
a(t)
<
+ K] +
+ K] + A + MG(Y))
(q'(t) +M[a'(t)]-) (q(t)[F(X)+K]+A+B+G(Y)).

q(t)
a(t)
a(t)
Integrating V'(t) from to to t, we get
and hence
V(t)
r (q, (s) +
ltD q(s)
Kl exp (1: ~gjdS)
< V (to) exp (
<
M [a' (s )]-)
= Kl
a(s)
qq(~~) ,
dS)
Kl is a constant.
Chapter 1
6
The boundedness of x(t) now follows as in Theorem 1.1.2.
If M
< 1, we define
q(t)
V(x, y, t) = a(t) [F(x)
+ K] + G(y) + M (A + B).
Then since M < 1, we have
V'(t)
::;
~~j (~~!~[F(X)+K]+ ~(A+B)+G(Y))

+ [a'(t)]a(t)
<
( q, (t)
q(t)
(q(t) [F(x)
a(t)
+ K] + A + MG(Y))
+ [a'(t)]-) V(t).
a(t)
The remainder of the proof follows as before.
Now, from Theorems 1.1.2 and 1.1.3 we deduce the following corollary.
Corollary 1.1.1. If in addition to the hypotheses of either Theorem 1.1.2
or 1.1.3, q(t)::; q2, q2 is a constant, and G(y) ---+ 00 as Iyl ---+ 00, then
all solutions of the system (1.1.2) are bounded.
Proof. From the proof of Theorem 1.1.2, we have
~~j
V'(t) ::;
so
V(t) ::; V(to) +
to
[: + V(t)] ,
q'(s)
( q(t) )
q to
m
n
-(-) V(s)ds + -In -(-) .

q
Hence, it follows that
V(t) ::; K2 exp
(1~ ~(~j dS)
::; K2
qq(~~)
<
00,
K2 is a constant.
The boundedness of y(t) now follows from that of G(y(O)).
The following lemma will be used to prove the convergence to zero of the
nonoscillatory solutions of equation (1.1.1). For this, we need the following
additional assumptions:
(Id xf(x) > 0 if x =I- 0 and f(x) is bounded away from zero if x is
bounded away from zero,
(h) condition (1.1.3) holds and r(t)/q(t)
---+
0 as t
---+ 00,
Preliminaries
(13) if x is bounded, then there exist a continuous function k(t~ and

to such that Ih(t, x, y)1 ::; k(t)g(y) for (t, x, y) E [iI, (0) x lR and
k(t)/q(t) --> 0 as t --> 00,
(14) g(y) ;:::: C > 0, c is a constant, ft': q(s)ds = 00 and ft': ds/a(s) = 00.
tl ;::::
Lemma 1.1.2. If conditions (h) - (14) hold and x(t) is a bounded

nonoscillatory solution of equation (1.1.1), then lim inft~oo Ix(t) I = O.
Proof. Let x (t) be a bounded nonoscillatory solution of equation (1.1.1),

say, 0 < x(t) < B for t;:::: T;:::: to, B is a constant, and let k(t) and
tl ;:::: T be determined by (h). Suppose that
liminfhoo x(t) '" O. Then
there exists t2 ;:::: tl sueh that x(t) is bounded away from zero for t;:::: t2'
Hence by condition (h), f(x(t));:::: A > 0 for t;:::: t2, A is a constant.
Choose t3;:::: t2 so that
r(t) < ~A
cq(t)
4
and
k(t) < ~A
q(t)
4
From equation (1.1.1), we have
(a( t ) x')'
e(t,x,x')
h(t,x,x')
() ( )
g(x') g(x') - q t f x
g(x')
<
-r(t) - k(t) - Aq(t)

c
< q(t) [~r(t)

c q(t) _
A]
Thus, (a(t)x')'::; -Acq(t)/2 < 0 for

from t3 to t, we find
a(t)x'(t) ::; a(t3)x'(t3) -
<
f
-"21 Aq()tort;::::
t3
t;:::: t3. Integrating this inequality
~Aclt q(s)ds
2
-->
00
as
--> 00,
t3
so there exists a t4;:::: t3 such that x' (t)
< 0 for t;::::
contradicting the fact that x(t) > 0 for

if x(t) < 0 for t;:::: T.
t;:::: T. A similar argument holds
t4' Hence,
The following two examples illustrate that condition (h) above is essential.
Example 1.1.1. Consider the equation
xl/(t)
+ tx'(t) + -x(t)
t
t> O.
(1.1.10)
Chapter 1
Here g(x') = 1 and h(t, x, x') = tx', and it is clear that all the hypotheses
of Lemma 1.1.2 are satisfied except (13) since we do not have Ih(t, x, x')1 :::::
k(t)g(x'). Equation (1.1.10) has a bounded nonoscillatory solution x(t) =
(t + l)/t which does not have lim inft-+oo x(t) = o.
Exrunple 1.1.2. The equation
satisfies all the conditions of Lemma 1.1.2 except (h). Here Ih(t, x, x')1 =
tlx'i ::::: t[l + (x'?] = tg(x') but k(t)/q(t) = t 2 f+ 0 as t ---> 00. Again
x(t) = (t + l)/t is a bounded nonoscillatory solution of equation (1.1.11).
In what follows it will be convenient to have the following notation.
Condition C. If x(t) is a nonoscillatory solution of equation (1.1.1),
then limt-+oo x(t) = O.
Also, we define
(t [q'(s)]- )
p(t) = exp ( - lto q(s) ds
and
b(t) = exp (-
lto
[a'(s)]a(s)
dS) .
Note that p(t)::::: 1 and b(t)::::: 1 for t 2: to.

Theorem 1.1.4.
hold,
Suppose conditions (1.1.3), (1.1.4), (1.1.8) and (1.1.9)
1
1
[q'(s)]ds <
q()
S
00
to
00
to
r(s) ds <
q(s)
(1.1.12)
00,
00
(1.1.13)
and there is a positive constant N such that

(1.1.14)
Then all solutions of equation (1.1.1) are bounded. If, in addition (II) (14) are satisfied, then condition C holds.
Proof. From (1.1.8), F(x) 2: -K for some constant K>
V(x, y, t) = b(t)p(t) (att) [F(x)
o.
+ K] + qtt) G(Y)) ,
Let
Preliminaries
then
,
{-a'(t)
V (t) = b(t)p(t) a2(t) [F(x)
+ K]-
q'(t)
a'(t) y2
q2(t) G(y) - a(t)q(t) g(y)-
- a(t)lq(t) h(t, x, y) (gryJ + a(t)lq(t) e(t, x, y)
(g~)
_(_1_
[F(x) + K] + _1 G(Y) ([a'(t)]- + [q'(t)]-)}
a(t)
q(t)
a(t)
q(t)
:S b(t)P(t){- [a'(t)]+ [F(x)+K]- [q'(t)]+ G(y) _ [q'(t)]- [F(x)+K]
a2(t)
[a'(t)]- G
- a(t)q(t) (y)
b
:S (t)p(t)
q2(t)
[a'(t)]- ( y2 )
g(y)
+ a(t)q(t)
a(t)q(t)
+ a(t)q(t) e(t, x, y)
Y )}
g(y)
{[a'(t)]- ( y2 )
r(t)
Iyl}
a(t)q(t) g(y) + a(t)q(t) g(y) .
Now conditions (1.1.9) and (1.1.12) imply that q(t) ~ ql > 0, p(t) ~ PI >
0, a(t) ~ al > 0 and b(t) ~ bl > 0 where ql, PI, al and bl are
constants. Also, Iyll g(y) :S NI for all y and some constant NI > O.
Thus, on integrating V'(t) from to to t, we get
V( t )
for all t
<
-
V()
to
+ -Nit
ql
to
[a'(s)]- d
NI i t r(s)d
s+-- s:S K I <
a(s)
al to q(s)
00
to, KI is a constant. Hence,

F(x(t
a(t)
a2
:S KI b(t)p(t) :::; KI blPI
for
t ~ to,
and so by condition (1.1.8), x(t) is bounded.

Next, let x(t) be a nonoscillatory solution of equation (1.1.1). Note
that by (II) we can choose K = O. Since lim inft-> 00 x(t) = 0 by Lemma
1.1.2, if x(t) is ultimately monotonic, we are done. If x(t) is not ultimately
monotonic, let E> 0 be given and choose tl ~ to so that (13) is satisfied
for t~tl' y(tl) =0,
and
10
Chapter 1
Then integrating V'(t) from t1 to t, we find for t::::: t1,
F(x(t
a2
q(t)
::; b(t)p(t) V(t) ::; b1P1 V(t)
a2 N
< -a2- V()
t1 + -
b1P1
This implies that

only if x(t) --> O.
b1P1 q1
it
limt-->oo x(t)
tl
[a'(s)J- dS
a( s)
1
+ -a2-Na1 b1P1
it
tl
r(s) ds
-<
q( s)
= 0, since by (Id, F(x(t
-->
t.
0 if and
Finally, we state the following two theorems.

Theorem 1.1.5.
Suppose conditions (1.1.3), (1.1.4), (1.1.7), (1.1.8),
(1.1.12) and (1.1.13) hold, and there is a positive constant L such that
Iyl/g(y) ::; L. Then all solutions of equation (1.1.1) are bounded. If in
addition (11) - (14) hold, then condition C holds.
Theorem 1.1.6.
Suppose conditions (1.1.3), (1.1.4), (1.1.6), (1.1.8),
r(s)ds/q1/2(s) < 00.
(1.1.9) and (1.1.12) hold, g(y)::::: c > 0 and
Then all solutions of equation (1.1.1) are bounded. If in addition (Id - (14)
hold, then condition C holds.
It:'
1.2. Some Useful Results from Analysis

and Fixed Point Theorems
In this section, we state some fixed point theorems and other results from
analysis which are needed throughout this book. We recall that a vector
space X equipped with a norm 11.11 is called a normed vector space. A
subset E of a normed space X is said to be bounded if there exists a
number K such that IIxll::; K for all x E E. A subset E of a normed
vector space X is called convex if for any x, y E E, ax + (1 - a)y E E
for all a E [O,lJ. A sequence {x n } in a normed vector space X is said
to converge to the vector x in X if and only if the sequence IIx n - xII
converges to zero as n --> 00. A sequence {x n } in a normed vector
space X is called a Cauchy sequence if for every E > 0 there exists an
N = N(E) such that for all n, m ::::: N(E), IIx n - xmll < E. Clearly a
convergent sequence is also a Cauchy sequence, but the converse may not
be true. A space X where every Cauchy sequence of elements of X
converges to an element of X is called a complete space. A complete
normed vector space is said to be a Banach space. Let E be a subset of
a Banach space X. A point x E X is said to be a limit point of E if
there exists a sequence of vectors in E which converges to x. We say
a subset E is closed if E contains all of its limit points. The union of
E and its limit points is called the closure of E and will be denoted by
E. Let X, F be normed vector spaces, and E be a subset of X. An
Preliminaries
11
operator T: E ----; F is continuous at a point x E E if and only if for any

E> 0 there is a 0> 0 such that IITx - Tyll < E for all y E E such that
Ilx - yll < 0. Further, T is continuous on E, or simply continuous, if it
is continuous at all points of E.
We say that a subset E of a Banach space X is compact if every
sequence of vectors in E contains a subsequence which converges to a
vector in E. We say that E is relatively compact if every sequence of
vectors in E contains a subsequence which converges to a vector in X,
i.e., E is relatively compact if E is compact.
A family E in C([a, b], JR) is called uniformly bounded ifthere exists
a positive constant K such that If(t)l:S; K for all t E [a, bJ and all
fEE. Further, E is called equicontinuous, if for every E > 0 there
exists a 0 = O(f) > 0 such that If(t l ) - f(t 2 )1 < E for all tl, t2 E [a, bJ
with It I - t21 < 0 and all fEE. A set E in C([a, 00), JR) is said to
be equiconvergent at 00 if all fEE are convergent in JR at the point
00, and moreover, for every E > 0 there exists a T ~ a such that for all
fEE, t ~ T =? If(t) -lims->oo f(8)1 < E.
Theorem 1.2.1 (Arzela-Ascoli Theorem). A subset E in C([a, b], JR)

with norm Ilfll = SUPtE[a,bJ If(t)1 is relatively compact if and only if it
is uniformly bounded and equicontinuous on [a, bJ.
Let B([T, 00), JR) be the Banach space of all continuous and bounded
real valued functions on the interval [T, 00 ), endowed with the usual
supnorm 11.11. The following result provides a compactness criterion for
subsets of B([T, 00), JR).
Compactness Criterion. Let E be an equicontinuous and uniformly
bounded subset of the Banach space B([T, 00), JR). If E is equiconvergent
at 00, it is also relatively compact.
Theorem 1.2.2 (Schauder's Fixed Point Theorem). Let E be a
closed, convex and nonempty subset of a Banach space X. Let T: E ----; E
be a continuous mapping such that T E is a relatively compact subset of
X. Then, T has at least one fixed point in E, i.e., there exists an x E E
such that Tx = x.
Theorem 1.2.3 (Schauder-Tychonov Fixed Point Theorem). Let
X be a locally convex linear space, S be a compact convex subset of X
and let T: S ----; S be a continuous mapping with T(S) compact. Then,
T has a fixed point in S.
Finally, we state the Lebesgue dominated convergence theorem.
Theorem 1.2.4 (Lebesgue's Dominated Convergence Theorem).
Let E be a measurable set and let {In} be a sequence of measurable
12
Chapter 1
functions such that limn->oo fn(x) = f(x) a.e. in E, and for every
n E N, Ifn(x)1 ::; g(x) a.e. in E, where g is integrable on E. Then
lim r fn(x)dx
n->ooJ
E
JrE f(x)dx.

1. The results of Section 1.1 are taken from Graef and Spikes [8]. For
several other related works, we refer to Burton and Grimmer [2], Grace and
Lalli [3-5], and Graef and Spikes [6-8].
2. For recent contributions in fixed point theorems and their applications see the recent monographs by Agarwal, Meehan and O'Regan [1], and
O'Regan and Precup [9].
1.4. References
1. R.P. Agarwal, M. Meehan and D. O'Regan, Fixed Point Theory and
Applications, Cambridge University Press, Cambridge, 200l.

2. T .A. Burton and R. Grimmer, On the asymptotic behavior of solutions
of x" + a(t)f(x) = e(t), Pacific J. Math. 41(1972), 43-55.
3. S.R. Grace and B.S. Lalli, On boundedness and asymptotic behavior
of certain second order integro-differential equations, J. Math. Phys. Sci.
14( 1980), 191-203.
4. S.R. Grace and B.S. Lalli, On continuability and boundedness of solutions of certain second order integro--differential equations, Indian J. Pure
Appl. Math. 12(1981), 950-963.
5. S.R. Grace and B.S. Lalli, Oscillation and convergence to zero of solutions of damped second order nonlinear differential equations, J. Math.
Anal. Appl. 102(1984), 539-548.
6. J.R. Graef and P.W. Spikes, Continuability, boundedness and asymptotic behavior of solutions of x" +q(t)f(x) = r(t), Ann. Mat. Pura Appl.
101(1974), 307-320.
7. J.R. Graef and P.W. Spikes, Asymptotic behavior of solutions of a
second order nonlinear differential equation, J. Differential Equations 17( 19
75), 461-476.
8. J.R. Graef and P.W. Spikes, Boundedness and convergence to zero of
solutions of a forced second order nonlinear differential equation, J. Math.
Anal. Appl. 62(1978), 295-309.
9. D. O'Regan and R. Precup, Theorems of Leray-Schauder Type and
Applications, Gordon & Breach, Amsterdam, 2001.
Chapter 2
Oscillation and Nonoscillation
of Linear Ordinary Differential
Equations
2.0. Introduction
The oscillation and nonoscillation property of solutions of second order
linear differential equations is of special interest, and therefore, it has been
the subject of many investigations. The interest in second order linear oscillations is due, in a large part, to the fact that many physical systems are
modelled by such equations. In this chapter we shall discuss some of the
most basic results in the theory of oscillations of linear ordinary differential equations of second order. In Section 2.1, we shall present Sturm and
Sturm-Picone comparison theorems which are useful in oscillation theory.
In Section 2.2, we shall provide some necessary and sufficient conditions for
the nonoscillation as well as some comparison theorems of Sturm's type.
Sufficiency criteria for the nonoscillation are given in Section 2.3. In Section 2.4, we shall establish sufficient conditions for the oscillation of second
order differential equations with alternating coefficients. Integral averaging
techniques as well as interval criteria for the oscillation are discussed in Section 2.5. In Section 2.6, several criteria for oscillation of linear second order
differential equations with integrable coefficients are established. Finally,
in Section 2.7 we shall discuss the problems of forced oscillations.
2.1. Sturm Comparison Theorem

This basic result is usually stated as follows:
Theorem 2.1.1. Let x(t) and y(t) be nontrivial solutions of
x"(t)
+ q(t)x(t)
= 0
and
y"(t)
+ ql(t)y(t)
= 0
Chapter 2
14
respectively in an interval [tl, t2] ~ JR. Suppose x(tt} = X(t2) = 0,

q(t), ql(t) E C([tl' t2],JR), and satisfy ql(t)::::: q(t), (ql(t) # q(t in
[tl,t2]' Then, y(t) must change sign in (h,t2)'
The standard method of proof is based on a 'Wronskian argument'
[4,29]' i.e., since
:t(y(t)x'(t) - x(t)y'(t = y(t)x"(t) - x(t)y"(t) = (ql(t) - q(tx(t)y(t),

it follows that
t2
y(t)x'(t) - x(t)y'(t)l t2 =
t,
t,
[ql(S) - q(s)]x(s)y(s)ds,
from which one derives the desired contradiction to the assumption that
x(t) > 0 and y(t) > in (tl, t2)'
This result extends readily to the solutions of
(a(t)x'(t)'
+ q(t)x(t) =
and
(a(t)y'(t'
+ ql(t)y(t) =
0,
where a(t) E C I ([tl,t2],JR+) and q(t), ql(t) E C([tl,t2],JR). In fact, to

prove this we only need the Sturmian identity
dt (y(t)a(t)x'(t) - x(t)a(t)y'(t
(ql(t) - q(tx(t)y(t).
If however, we wish to compare the solutions of
and
(a(t)x'(t'
+ q(t)x(t) =
(2.1.1)
(al(t)y'(t'
+ ql(t)y(t) =
0,
(2.1.2)
where a(t), al(t) E C I ([tl,t2], JR+) and q(t), ql(t) E C([tb t2], JR) under
the hypotheses
ql(t) ::::: q(t) (ql(t)
# q(t
in [h, t2], a(t) ::::: al(t),
(2.1.3)
then the Sturmian identity which takes the form
:t
(y(t)a(t)x'(t) - x(t)al(t)y'(t
(ql(t) - q(tx(t)y(t)
+(a(t) - al(tx'(t)y'(t),
is not useful since the signs of x'(t) and y'(t) are not known. However,
if y(t) # 0, then this identity modified by Picone [26] becomes
! (:g~
(y(t)a(t)x'(t) - x(t)al (t)y' (t))) = (ql(t) - q(tx(t)y(t)
+ (a(t) - al(t(x'(t2
+ al(t)
(x'(t) _ :i:?y'(t) 2,
(2.1.4)
Oscillation and non oscillation of linear ordinary differential equations
15
which immediately yields a more general result. Indeed, if X(tI) = X(t2) =

0, yet) >
in [tI, t2] and (2.1.3) holds in [tI, t2], we can integrate
(2.1.4) between tl and t2 to obtain the desired contradiction.
Thus the Picone identity (2.1.4) provides an elementary, but rigorous
proof of Sturm's original theorem. This result is known as Sturm-Picone
theorem which we state in the following:
Theorem 2.1.2. If there exists a nontrivial real-valued solution x(t) of
equation (2.1.1) in (tI' t2) such that x(td = x(t 2 ) = 0, and
t2
[(a(s) _ aI(s))(x'(s))2
t,
+ (qI(S)
- q(s))x2(s)] ds 2:
then every solution yet) of (2.1.2) has one of the following properties:
yet) has at least one zero in (tI' t2), or

(ii) yet) is a constant multiple of x(t) on [tI,t2].
(i)
2.2. Nonoscillatory Characterizations

and Comparison Theorems
In this section, we shall establish necessary and sufficient conditions for
the nonoscillation of equation (2.1.1) when aCt) E C([to, 00), R+) and
q(t) E C([to, 00), R) for some to 2: O. For this, we shall need the following
lemma.
Lemma 2.2.1. Equation (2.1.1) is nonoscillatory if and only if there exists
a function u(t) E CI([to,oo),R) which satisfies the inequality
u'(t)
u 2 (t)
aCt)
+ q(t)
eventually.
Consider the more general equation
(a(t)x'(t))'
+ p(t)x'(t) + q(t)f(x(t)) =
0,
(2.2.1)
where aCt) E CI([to,oo),R+), pet) E CI([to,oo),R), q(t) E C([to,oo),R)

and f E C(R, R). We shall assume that
(id xf(x) > 0 for x =I- 0, and

(i2) f'(x) 2: k > for x =I- 0, where k is a real constant.
Now, we are in the position to prove the following nonoscillation result.
Chapter 2
16
Theorem 2.2.1. Assume that there exists a function p(t) E C2([to, 00),
JR+). If (2.2.1) is nonoscillatory, then the equation
(a(t)p(t)y' (t'
+ kQ*(t)y(t)
(2.2.2)
is also nonoscillatory, where
Q*(t)
p(t) [q(t) + ka(t)h*2(t) - (a(t)h*(t' - (p(t)h*(t]
and
h*(t)
(2.2.3)
= p(t)p(t) - a(t)p'(t).
(2.2.4)
2 k a(t)p(t)
Proof. Let x(t) be a nonoscillatory solution of (2.2.1). Define
= p(t)a(t)
w(t)
[f~~~t;
+ h*(t)] , t:::: to::::
0.
Then, we have
w'(t)
= p(t) [(a(t)x'(t' +(a(t)h*(t'- a(t)f'(X(t(X'(t2] + p'(t) w(t)

f(x(t
j2(x(t
:S -p(t)q(t) - p(t)p(t)
f~~~~) + p(t)(a(t)h*(t' -
*) (a(t)p(t (f(x(t
x'(t)
+ ( P(t)
a(t) - 2kh (t)
= -Q*(t) - ka(t)p(t)
or
w'(t)
(f~~~;
+ h*(t) 2
+ a(t)p(t) w 2 (t) + Q*(t) :S
+h
ka(t)p(t)
*)
(t)
= -Q*(t) -
0,
p(t)
(f~S;)
a(t~(t) w 2(t),
eventually.
Applying Lemma 2.2.1, we arrive at the desired conclusion.
Corollary 2.2.1. If there exists a function p(t) E C2([to, 00), JR+) such
that equation (2.2.2) is oscillatory, where the functions Q*(t), h*(t) are
defined in (2.2.3) and (2.2.4) respectively, then (2.2.1) is oscillatory.
Remark 2.2.1.
defined as
From (2.2.4), it is clear that the function p(t) can be

(2.2.5)
For the special case when f(x) = x and p(t) = 0, i.e., when (2.2.1)
is reduced to equation (2.1.1), we have the following interesting result.
17
Lemma 2.2.2. Let p(t) E C2([to,00),JR+) be a given function. Then,

(2.1.1) is oscillatory if and only if the equation
(a(t)p(t)w'(t))'
+ Q(t)w(t)
(2.2.6)
= 0
is oscillatory, where
Q(t)
p(t) [q(t)
+ a(t)h 2(t)
and
h(t)
- (a(t)h(t))']
(2.2.7)
p'(t)
(2.2.8)
= - 2p(t)'
Combining Theorem 2.1.2 and Lemma 2.2.2, we obtain the following

result.
Theorem 2.2.2. Let a(t), al(t) E Cl([tl' t2], JR+) and q(t), ql(t) E
C([tI, t2], JR). Suppose the boundary value problem
(a(t)x'(t))'
+ q(t)x(t)
0,
x(td
X(t2)
has a solution x(t) with x(t) -=I 0 on (tl' t2) and there exist two functions
p(t), PI (t) E C 2([tl' t2], JR+) such that
t2
{[a(s)p(s) - al(s)pds)](x'(s))2
+ (Ql(S) -
Q(s))x 2(s)} ds ::::: 0,
t\
where the functions Q, h are defined in (2.2.7) and (2.2.8) respectively,

(2.2.9)
and
(2.2.10)
Then every solution y(t) of equation (2.1.2) must have a zero in (tl' t2)
unless x(t) and y(t) are proportional.
Corollary 2.2.2. Let a(t), adt) E Cl([to, 00), JR+) and q(t), ql(t) E
C([to, (0), JR). Suppose there exist two functions p(t), PI (t) E C2([to, (0),
JR+) such that al(t)Pl(t) S; a(t)p(t) and Ql(t)::::: Q(t) on [to,oo),
where the functions Q, Ql are defined in (2.2.7) and (2.2.9) respectively.
If equation (2.1.1) is oscillatory, then (2.1.2) is also oscillatory.

Example 2.2.1. Consider the Euler differential equation
x"(t)
+ t~x(t) =
(2.2.11)
and the differential equation
(2.2.12)
18
Chapter 2
on the interval [1,00), where C and Cl are real constants. We know that
the Euler equation (2.2.11) is oscillatory if c > 1/4 and nonoscillatory
if c:-::; 1/4. Let p(t) = 1 and Pl(t) = 1/t2 for equations (2.2.11) and
(2.2.12) respectively. Then it follows from Corollary 2.2.2 that (2.2.12) is
oscillatory if Cl 2: C> 1/4 and nonoscillatory if Cl :-::; C :-::; 1/4.
Now, in what follows we shall assume that the functions p(t), PI (t) E
C([to, 00), lR+), Q, h, Ql and hI are defined in (2.2.7), (2.2.8), (2.2.9)
and (2.2.10) respectively,
00
/
-,-,.---..,.-ds
a(s)p(s)
00
(2.2.13)
and for fixed t 2: to,
(3(t) =
00
Q(s)ds < 00,
(31(t) =
00
Ql(s)ds < 00.
(2.2.14)
Theorem 2.2.3. The following four statements are equivalent:
(i) Equation (2.1.1) is nonoscillatory.

(ii) There exist a number T 2: to and a function v(t) E C([Tl,OO),lR)
such that
00
v2(S)
(2.2.15)
v(t) = (3(t) + t a(s)p(s) ds for t 2: T.
In particular, if w(t) is a nonoscillatory solution of equation (2.2.6), then

v(t) can be taken as v(t) = a(t)p(t)w'(t)/w 2 (t) for t 2: T.
(iii) There exist a number T 2: to and a function v(t) E C([T,oo),lR)
such that
(2.2.16)
(iv) There exists a function y(t) E C1([T,00),lR) for some T 2: to such
that for t 2: T
,
y2 (t)
(2.2.17)
Y (t) + Q(t) + a(t)p(t) :-::; O.
Proof (i) => (ii). Since equation (2.1.1) is
nonoscillatory by Lemma 2.2.2. Let w(t) be
any loss of generality, we may assume that
Define v(t) = a(t)p(t)w'(t)/w(t) for t 2: T.
,
V
(t)
v 2 (t)
+ Q(t) + a(t)p(t)
= 0
nonoscillatory, (2.2.6) is also

a solution of (2.2.6). Without
w(t) > 0 for t 2: T 2: to.
Then, we have
for t 2: T.
Oscillation and nonoscillation of linear ordinary differential equations
19
Integrating the above equation from t to u, u:::: t :::: T, we find
v(u) - v(t)
We claim for fixed
lu
t:::: T
lu
Q(s)ds +
v 2 (s)
a(s)p(s) ds = O.
that
00
v2(s)
--:--:-'--:--:-ds <
a(s)p(s)
For this, assume that JtOO v 2(s)j(a(s)p(s))ds

there is a T 1 :::: T such that
v(u) for
r
iT1
u::::: T 1 ,
v 2 (s)
a(s)p(s) ds
(2.2.18)
= v(t) -
(2.2.19)
00.
Q(s)ds -
00
IT1
t
for some t :::: T, then

v 2 (s)
a(s)p(s) ds < - 1
or equivalently,
(2.2.20)
It follows that w'(u)
< 0 for u:::: Tl and
v 2(u)
a(u)p(u)[l
v(u)
+ J;1 a(s2)~(~)dsl
> - --:--:'-':-:-
a(u)p(u)
w'(u)
w(u)
---
for
u:::: T 1 .
Integrating the above inequality from Tl to u, we get

In(l+
h~ a~2)~ls)dS)
::::
In(:~;/)
for
u::::T1 .
Using (2.2.20) in the above inequality, we obtain -w' (u) :::: w(Td(a(u)p(u))
for u:::: T 1 . Finally, integrating this last inequality, we find that w(u)----+
-00 as u ----+ 00, which contradicts the fact that w(t) > 0 for t:::: T.
Thus, (2.2.19) holds.
Now, letting u ----+ 00 in (2.2.18), it follows that v(u) ----+ v(oo) < 00.
Hence, in view of (2.2.19), v(oo) = O. This in (2.2.18) now implies (2.2.15),
i.e., (ii) holds.
(ii) => (iii). It is obvious.
(iii) => (iv). Suppose there exists a function v(t) E C([T, (0), JR) satisfying
(2.2.16). Define y(t) = (3(t) + JtOO v 2(s)j(a(s)p(s))ds. Then, Iv(t)l::::
ly(t)l, and
y' (t)
y2(t)
v 2 (t)
- Q(t) - a(t)p(t) < - Q(t) - a(t)p(t)'
20
Chapter 2
Hence, (iv) holds.

(iv) =} (i). By Lemma 2.2.2, equation (2.2.6) is nonoscillatory and hence
(2.1.1) is nonoscillatory. This completes the proof.

Next, we present two nonoscillatory characterizations of (2.1.1).
Theorem 2.2.4. The following three statements are equivalent:
(i) Equation (2.1.1) is nonoscillatory.

(ii) There exist T~ to and a function y(t) E C([T,oo),JR) such that
y(t)
~ ~(t) +
1= a~;;;~~)
y2(s)ds
for
~ T.
(2.2.21)
(iii) There exist T 2' to and a function z(t) E C([T, (0), JR) such that
z(t)
where
~(t) +
1 /-*
1
00
~(t) =
t]
a(s);(s) z2(s)ds
2' T,
f32(s)
a(s)p(s) M[S, t]ds
(1
and
for t
M[S, t] = exp 2
f3(T)
)
a(T)p(T) dT .
(2.2.22)
(2.2.23)
(2.2.24)
Proof (i) =} (ii).

Assume that equation (2.1.1) is non oscillatory. It
follows from Theorem 2.2.3 that (2.2.15) has a solution v(t). Define y(t) =
ft= v 2(s) / (a( s )p( s) )ds. Then, y' (t) = -[y(t) + f3(t)f / (a(t)p(t)), or
,2f3(t)
y2(t)
f32(t)
+ a(t)p(t) y(t) = - a(t)p(t) - a(t)p(t)'
Y (t)
(2.2.25)
Using M[S, t] as an integrating factor and integrating (2.2.25) from t to

u 2' t 2' T, we get
(2.2.26)
The left-hand side of inequality (2.2.26) is independent of u, and hence
we may let u --> 00 in (2.2.26). This immediately leads to (2.2.21).
(ii) =} (iii). Suppose there exists a function y(t) E C([T, (0), JR) satisfying
(2.2.21). Define zo(t) = y(t), and inductively for n = 1,2""
(2.2.27)
21
Using (2.2.21) and (2.2.27), it is easy to show by induction that 0:::; ~(t) :::;
zn(t) :::; Zn-l(t) :::; yet). Thus, the sequence {zn(t)} has a pointwise
limiting function z(t) = limn~oo zn(t). Since the integrand in (2.2.27) is
nonnegative, it follows from the Lebesgue dominated convergence theorem
that z(t) satisfies the equation (2.2.22).
(iii) =? (i).
Assume that (2.2.22) has a solution z(t), then an easy
computation shows that vet) = z(t) + (3(t) defines a solution of (2.2.17).
As a consequence of Theorem 2.2.3, we conclude that equation (2.l.1) is
nonoscillatory.
(i) Equation (2.l.1) is nonoscillatory.

(ii) There exist T;:;' to and a function yet) E C([T,oo),lR) such that
yet) ;:;, C(t)
00
y2(s)
1l*[S, tj a(s)p(s) ds,
(2.2.28)
(iii) There exist T;:;' to and a function z(t) E C([T,oo),lR) such that
z(t) = C(t)
00
z2(S)
1l*[S, tj a(s)p(s) ds,
where
00
C(t)
exp
(2.2.29)
e(s)
f1*[s, tj a(s)p(s)ds
(2.2.30)
(2.2.31)
and
11 [s, t] =
t;:;,
(3(T) + ~(T) )
a(T)p(T) dT .
Proof (i) =? (ii).

Assume that equation (2.l.1) is nonoscillatory. It
follows from Theorem 2.2.4 that (2.2.22) has a solution z(t). Define yet) =
ftCJO Il[s, tjz2(s)/(a(s)p(sds. Then, we have
y'(t) =
- a(t)p(t) [y2(t)
+ e(t) + 2((3(t) + ~(ty(t)l
(2.2.32)
Using Il*[s, t] as an integrating factor and integrating (2.2.32) from t to

u ;:;, t ;:;, T, we get
yet)
Il*[u,t]y(u)+
>
e(s)
Il*[s,t j a (s)p(s)ds+
*
~2(s)
11 [s,tja(s)p(s)ds+
y2(S)
Il*[s,t j a (s)p(s)ds
*
y2(s)
11 [s,tja(s)p(s)ds.
(2.2.33)
22
Chapter 2
Letting u
-->
00 in (2.2.33), we obtain (2.2.28).
(ii) =} (iii). Suppose there exists a function y(t) E C( [T, 00), JR) satisfying
(2.2.28). Define zo(t) = y(t), and inductively for n = 1,2, ..
Zn (t ) = ~ * ()
t
00
*[
fi s, t
z;
1 ()-1 ((S )) ds.

asp s
(2.2.34)
Using (2.2.27) and (2.2.34), it is easy to show by induction that O:s: C(t) :s:
Zn(t) :s: zn-dt) :s: y(t). Thus, the sequence {zn(t)} has a pointwise
limiting function z(t) = lim n ..... oo zn(t). Since the integrand in (2.2.34) is
nonnegative, it follows from the Lebesgue dominated convergence theorem
that z(t) satisfies (2.2.29).
(iii) =} (i).
Assume that (2.2.29) has a solution z(t), then an easy
computation shows that v(t) = z(t) + (3(t) + ~(t) defines a solution of
(2.2.17). As a consequence of Theorem 2.2.3, we conclude that equation
(2.1.1) is nonoscillatory.
Remark 2.2.2. It is clear from Theorems 2.2.4 and 2.2.5 that the process of generating higher order iterated Riccati integral equations can be
continued.
Now, using above nonoscillatory characterizations of solutions of equation (2.1.1), we shall derive the following comparison results.
Theorem 2.2.6. With respect to equations (2.1.1) and (2.1.2) assume
that there exist two functions p( t), pdt) E C 2([ to, 00), lR+) such that
(2.2.35)
If (2.1.2) is nonoscillatory, then equation (2.1. 1) is also nonoscillatory.
Proof. Since equation (2.1.2) is nonoscillatory, it follows from Theorem
2.2.3 that there exist T 2: to and a function v(t) E C([T,oo),lR) such
that
Iv( t)1
2:
1(31 (t) +
00
a1
~2)~:(S) ds 1
for
t 2: T.
(2.2.36)
Using (2.2.35) in inequality (2.2.36), we obtain
Iv(t)1
2:
1(3(t) +
1= a~s~~ls)
dsl
for
t 2: T.
Now, it follows from Theorem 2.2.3 that (2.1.1) is nonoscillatory.

Example 2.2.2.
Consider the differential equation
() = 0,
x /I (t) + a sin
t It
xt
(2.2.37)

where a and "I
i:- 0
are constants, and a nonoscillatory Euler's equation

1
+ 4t 2y(t) =
y"(t)
Let
p(t) = exp
Then as t
(2~
---4
It
23
co: "Is
dS),
17 =
t 2: 1.
0,
2~1
(2 + sgn
(2.2.38)
~)
and pdt)
1.
00, p(t) = 1 + 0 (lit) 2: 1 = PI (t) and
Thus, we have
as
---4
00
1
and (3I (t) = - .
4t
It follows from Theorem 2.2.6 that equation (2.2.37) is nonoscillatory if
Inhl < I/V2

that there exist two functions p(t), PI(t) E C 2 ([to,00),lR+) such that
a(t)p(t) 2: al (t)PI (t),

and
(32 (t)
a(t)p(t)
(3f(t)
<
al (t)PI (t)
(3(t)
(3I (t)
<
a(t)p(t)
al (t)PI (t)
for
for
t 2: to
t 2: to
(2.2.39)
(2.2.40)
If (2.1.2) is nonoscillatory, then equation (2.1.1) is also nonoscillatory.

2.2.4 that there exist T 2: to and a function y(t) E C([T,oo),lR) such
that
y(t) = 6(t)
OO
MIls, t] 2
() ()y (s)ds,
al s PI s
where
6(t) =
(2.2.41)
and
Mds, t]
(2.2.42)
24
Chapter 2
From (2.2.39) and (2.2.40), we have 6(t)::::: ~(t) and p,Ils, t] ::::: p,[s, t],
where ~(t) and p,[s, t] are defined in (2.2.23) and (2.2.24) respectively.
Hence,
y(t) ::::: ~(t)
00
p,[s t]
a(s);(s)y2(s)ds.

that there exist two functions p( t), PI (t) E c 2([ to, 00), JR+) such that
~2(t)
<
~i(t)
a(t)p(t) and
al (t)Pl (t)
,8(t) + ~(t) < ,81 (t) + 6 (t)

a(t)p(t) al(t)pr(t) '
(2.2.43)
(2.2.44)
where ~(t), 6(t) and p,Ils, t] are defined in (2.2.23), (2.2.41) and
(2.2.42) respectively. If (2.1.2) is nonoscillatory, then equation (2.1.1) is
also nonoscillatory.
2.2.5 that there exist T::::: to and a function y(t) E C([T, 00), JR) such
that
00
*
*
y2(S)
y(t) = ~1 (t) +
p,Ils, t] () () ds,
t
al s PI s
where
~;(t)
and
p,i[s,t] = exp
1
1
00
~i(s)
() ()P,ds,t]ds
al s PI s
(218 (/3r(T)+6(T))
dT).
al(T)p(T)
t
From (2.2.43) and (2.2.44), we have ~i(t)::::: C(t) and p,ils, t] ::::: p,*[s, t],
where C(t) and p,*[s, t] are defined in (2.2.30) and (2.2.31) respectively.
Hence,
y(t) ::::: C(t)
1 ~:~~(;)
00
y2(s)ds,
t::::: T.
As an immediate consequence of the above theorems, we state the following result.

Corollary 2.2.3. With respect to equations (2.1.1) and (2.1.2) assume
that there exist two functions p(t), PI (t) E C 2 ([to, 00), JR+) such that
a(t)p(t) ::::: al (t)Pl (t), ~(t) ~ 6 (t) and ,8(t) + ~(t) ~ ,81 (t) + 6 (t), where
25
and 6(t) are defined in (2.2.23) and (2.2.41) respectively. If (2.1.2)

is nonoscillatory, then equation (2.1.1) is also nonoscillatory.
~(t)
Remark 2.2.3. It is clear from Theorems 2.2.7 and 2.2.8 that we can establish the higher order iterated comparison theorems by using the nonoscillatory characterizations.

In this section, we shall present easily verifiable nonoscillation criteria
for equation (2.1.1).
Theorem 2.3.1. Equation (2.1.1) is nonoscillatory if and only if there
exist T ~ to and a function h(t) E C 1 ([T,oo),lR) satisfying
q(t)
+ a(t)h 2(t) - (a(t)h(t' ::;
0 for
t ~ T.
Proof. Let x(t) be a nonoscillatory solution of equation (2.1.1) such that

x(t) -=/:- 0 for t ~ T ~ to. Define h(t) = -x'(t)/x(t) for t ~ T. Then it
follows from equation (2.1.1) that
q(t)
+ a(t)h 2(t) - (a(t)h(t'

()
qt
+a
( ) ( x'(t)2
t
x (t)
x 2 (t)
q(t) - q(t) x 2(t)
(a(t)x'(t' x(t) - aCt) (X'(t2

x 2 (t)
o.
Conversely, if q(t)+a(t)h 2(t) - (a(t)h(t' ::; 0, then it follows from Corollary 2.2.2 that (2.2.6) is nonoscillatory because equation (a(t)p(t)w'(t' =
is nonoscillatory, where pet) = exp

(-2 J~ h(s)ds).
Thus, equation
Example 2.3.1. Consider the Euler differential equation (2.2.11). Let

h(t) = -1/(2t). Then, h'(t) = 1/(2t2) and
2
,
c
1
1
q(t) + h (t) - h (t) = - 2 + - 2 - t
4t
2t2
4c-1
--;w-.
Thus, by Theorem 2.3.1, equation (2.2.11) is nonoscillatory if c::; 1/4.

Example 2.3.2. Consider the differential equation
x"(t)
[4~2 + (tl~t)2] x(t)
= 0,
(2.3.1)
26
Chapter 2
where c is a real constant. Let
Then, we have
and
1
4t 2
q(t) + h 2 (t) - h'(t)
1 (1
1(1
t2
4c -1
4(tlnt)2
-"2
1)2
+ 4"
t + tInt
1)
(tlnt)2
1
(tlnt)2
+ t 2 lnt +
c:::; 4
when
Thus, by Theorem 2.3.1, equation (2.3.1) is nonoscillatory if c:::; 1/4.

As applications to Theorem 2.3.1, we present the following two corollaries.
Corollary 2.3.1. If a'(t) :::; 0 for t
to, and
t 2 q(t)
hmsup - (-) < 4-'

t ..... oo
a t
(2.3.2)
then equation (2.1.1) is nonoscillatory.

Proof. It follows from (2.3.2) that there exist two numbers T ~ to and
c < 1/4 such that q(t):::; ca(t)/t 2 for t ~ T. Let h(t) = -1/(2t). Then,
h'(t) = 1/(2t 2 ) and
q(t)
+ a(t)h 2 (t)
4c -1
- (a(t)h(t))' :::; 4t;2a(t) - a'(t)h(t) :::; 0 for
t ~ T.
This and Theorem 2.3.1 imply that equation (2.1.1) is nonoscillatory.

Corollary 2.3.2. If a'(t) :::; 0 for t
limsup (tlnt)2 (qt))

t ..... oo
a t
to, and
-~)
4t
< -41 ,
(2.3.3)

Proof.
that
By (2.3.3), there exist two numbers T
q(t) < aCt)
(4~2 + (t l~ t)2 )
for
to and c < 1/4 such

t
~ T.
27
Let
h(t) =
- -1 (1-
Then for t 2: T, we have
h'(t) _
-
and
q(t)
(~
+ - 1) .
tIn t
_1_ + __1_)
(t t)2
t 2 + t 2 In t
In
+ a(t)h 2(t) - (a(t)h(t))'

< a(t)
[4~2 +(tl~t)2 +~ (~+ tl~tr
- ~ C~
+ t2 ~nt +
(tl~t)2)] -
a'(t)h(t)
a(t) ~~~:t)11- a'(t)h(t) :::; 0 for
2: T.
Now, it follows from Theorem 2.3.1 that equation (2.1.1) is nonoscillatory.
In what follows, we shall assume that p(t) E C 2 ([to, 00), JR+) and
h(t) = -p'(t)/(2p(t)). Clearly, from Lemma 2.2.2 and Theorem 2.3.1 the
following result is immediate.
exist T 2: to and a function v( t) E C 1 ([T, 00), JR) satisfying
Q(t)
+ a(t)p(t)v 2(t) - (a(t)p(t)v(t))' :::;
0 for
2: T,
where Q is defined in (2.2.7).

As applications to Theorem 2.3.2, we have the following corollaries.
Corollary 2.3.3. Suppose there exist two functions g(t) E C 1 ([to, 00),
JR+) and 'IjJ(t) E C 1([to,00),JR) such that g'(t) 2: 1/(a(t)p(t)) and
'IjJ'(t) :::; -Q(t) for t 2: to. If
1
limsup g(t)I'IjJ(t)1 < -4'
t->oo
(2.3.4)
then equation (2.1.1) is oscillatory.

Proof. By (2.3.4), there are two numbers T 2: to and k E (0,1/4) such
that 1'IjJ(t) I < k/g(t) for t 2: T. Let
v(t) =
1
[
- 2a(t)p(t) 2'IjJ(t)
1 - 2k]
+ g(i)
.
28
Chapter 2
Then, we have
(a(t)p(t)v(t))'
-'l//(t)
p(t) [q(t)
(1
~:2~~r(t)
+ a(t)h 2(t) - (a(t)h(t))'] + 2a (t~ p(~\(

)
t 9 t
and
p(t) [q(t)
+ a(t)h 2(t) - (a(t)h(t))'] + a(t)p(t)v'(t) - (a(t)p(t)v(t))'

1
::; 4a(t)p(t) 2V;(t)

1
[2
1 - 2k] 2
+ ---g(i)
4a(t)p(t) 4V; (t)
<
1 - 2k
)
2a(t)p(t)g2(t)
4(1-2k)
g(t) V;(t)
(1-2k)2
+ ---g(i)
2(1-2k)]
g2(t)
( ) (1 ) 2() [2
4k + 4k(1- 2k) + (1 - 2k) 2
- 2(1 - ]
2k)
4a t p t 9 t
4k -1
for t 2 T.
2 ( ) ::; 0
(
4a t ) p( t ) 9 t
For the special case of (2.1.1), namely, the equation
x"(t)
where q(t) E C([to, 00), JR+),
result.
+ q(t)x(t) =
Proof.
(2.3.5)
Corollary 2.3.3 reduces to the following
Corollary 2.3.4. If ftoo q(s)ds < 00 and

1/4, then equation (2.3.5) is nonoscillatory.
'l/J(t)
0,
lim SUPt-> 00 t ftOO q(s)ds
<
Apply Corollary 2.3.3 with a(t) = I, p(t) = I, g(t) = t and
= ftoo q(s)ds.
Similarly, we have the following corollary.

Corollary 2.3.5. Let g(t) E C 1 ([to, 00), JR+) and V;(t) E C 1([to,oo),JR).
Suppose g(t) and V;(t) satisfy g'(t) ::; -l/(a(t)p(t)) and v;'(t) 2 -Q(t)
for t 2 to. If limsuPt->oo g(t)Iv;(t)1 < 1/4, then equation (2.1.1) is
nonoscillatory.
Corollary 2.3.6. Let p(t) E C 2 ([to, 00), JR+), g(t) E C 1 ([to, 00), JR+)
and V;(t) E C 1 ([to,oo),JR). Suppose g(t) and V;(t) satisfy g'(t) 2
l/(a(t)p(t)) and v;'(t)::; -Q(t) for t 2 to. If there exists a number
c > 0 such that - y'c - c ::; g(t)V;(t) ::; y'c - c ::; 1/4, then equation

Proof. Let
v(t) =
- a(t)lp(t) ('lj;(t)
+ g~t))
29
Then, we have
(a( t )p( t )v( t))'
- 'lj;'(t)
+ c:~~~~ >
Q(t)
+ a(t)p(~)g2(t)
and
Q(t)
+ a(t)p(t)v 2(t)
- (a(t)p(t)v(t))'
< a(t)lp(t) ('lj;(t) +
g~t)) a(t)p(~)g2 (t)

2 -
(g2~t) - g2~t))
::::: a(t)lp(t)
0 for
2: to
> 0 such that
Corollary 2.3.7. If there exists a number c
and
ft
OO
q(s)ds <
00,
Proof. Apply Corollary 2.3.6 with p(t) = 1, g(t) = 1 + ft~) l/a(s)ds and
'lj;(t)
= f t q(s)ds.
OO
Example 2.3.3. Consider equation (2.3.1). Let p(t)

). < 1 is a constant. Then, we find
h(t)
1(1 ).)
-2"
and
t+tlnt
= t(lnt)>.,
where
Q(t)
Let
g(t) =
1
1 A
1 _ ). (In t) and 'lj;(t) =
Then, we have g'(t)
2), + 4c
A-I
4(1 _ ).) (In t)
.
).2 -
= l/(a(t)p(t)), 'lj;'(t) = -Q(t)
g(t)'lj;(t)
2), + 4c < ~
4(1 -).)2
- 4
).2 -
and
1
if c <-.
-4
Now, it follows from Corollary 2.3.6 that (2.3.1) is nonoscillatory if c:::::

1/4.
Chapter 2
30
A result similar to that of Corollary 2.3.6 is the following:

Corollary 2.3.8. Let p(t) E C 2([to,00),JR+), g(t) E C 1([to,00),JR+)
and 'IjJ(t) E C 1([to, 00), JR). Suppose g(t) and 'IjJ(t) satisfy g'(t) :s;
-l/a(t)p(t)) and 'Ij/(t) 2": -Q(t) for t 2": to. If there exists a number
c > 0 such that - vic - c :s; g(t)'IjJ(t) :s; vic - c :s; 1/4, then equation
Corollary 2.3.9. Let p(t) E C 2([to, 00), JR+), g(t) E C1([to, 00), JR+)
and 'IjJ(t) E C 1 ([to,00),JR). If limt-.oo'IjJ(t) exists, g'(t) 2": l/(a(t)p(t))
and 'Ij/(t):S; -g(t)Q(t) for t 2": to, then equation (2.1.1) is nonoscillatory.
Proof. Since limt->oo 'IjJ(t) exists, there are numbers T 2": to and m such
that 0 < m + 'IjJ(t) :s; 1 for t 2": T. Let v(t) = (m + 'IjJ(t))/(a(t)p(t)g(t))
for t 2": T. Then, we have
'IjJ'(t) >
g'(t)(m + 'IjJ(t))
g2(t)
- g(t)
Q(t)
(a(t)p(t)v(t))'
m + 'IjJ(t)
+ a(t)p(t)g2(t)
and for t 2": T,
Q(t)
+ a(t)p(t)v2 (t)
- (a(t)p(t)v(t))'
< (m+'IjJ(t))2 _
-
a(t)p(t)g2(t)
m + 'IjJ(t)
m+'IjJ(t)
< m + 'IjJ(t)
a(t)p(t)g2(t)- -a(:-:t)-p(:-:t)'-'g2"""(---:-t) - a(t)p(t)g2 (t) =
o.
Corollary 2.3.10. Let p(t) E C 2([to, 00), JR+), g(t) E C1 ([to, 00), JR+)
and 'IjJ(t) E C 1([to, 00), JR). If limt->oo 'IjJ(t) exists, g'(t):s; -l/(a(t)p(t))
and 'IjJ'(t) 2": -g(t)Q(t) for t 2": to, then equation (2.1.1) is nonoscillatory.
Next, we present the following results.

Corollary 2.3.11. Let p(t) E C 2([to, 00), JR+) and g(t) E C 1([to, 00),
JR+) satisfy Q(t):S; 1/g2(t) for t 2": to. Ifeither g'(t)-(l/(a(t)p(t))) 2": 1
for t 2": to, or
lim
t->oo
(g' (t) -
()1 ( ))
a t pt
m exists, m > 1 is a constant,

Proof. It follows from the hypothesis that there is a number T 2": to such
that g'(t)-(l/(a(t)p(t))) 2": 1 for t 2": T. Let v(t) = -l/(a(t)p(t)g(t)), t 2":
T. Then for t 2": T, we have
Q(t)
=
+ a(t)p(t)v(t) - (a(t)p(t)v(t))'
_1_
g2(t)
+ a(t)p(t)g2(t)
_ g'(t) = _1_ (1
g2(t)
g2(t)
+ a(t)p(t)
_, t ) < 0
9 ()

Now, by Theorem 2.3.2 equation (2.1.1) is nonoscillatory.
31
Example 2.3.4. Consider the differential equation (2.3.1) with c ~ 1/4.

Let p(t) = tin t for t 2: to > 1. Then for t 2: to, we have
h t
_ _ ~ (~
2p( t) 2 t
_ _ p'(t)
() -
_1_)
+ tin t
Q(t)
and
4c-1
4tlnt ~ O.
If g(t) = Hlnlnt, t> e, then Q(t) ~ 1/g2(t) and g'(t)-(l/(tlnt = 1.

Thus, by Corollary 2.3.11 equation (2.3.1) is nonoscillatory.
Corollary 2.3.12. Let p(t) E C 2([to,00),lR+) and 'lj;(t) E C 1 ([to, 00),

lR). Suppose 'lj;(t) satisfies 'lj;'(t) ~ -Q(t) for t 2: to. If
00
'lj;2(S)
1
a(s)p(s) ds ~ 4'lj;(t) ,

Proof. Let
1
- a(t)p(t) 'lj;(t)
v(t)
<
'lj;2 (S )
00
a(s)p(s) ds .
and
+ a(t)p(t)v 2(t) - (a(t)p(t)v(t'

1
- a(t)p(t)
1
= a(t)p(t)
+ 4'lj;2(t)/(a(t)p(t
Then, we have (a(t)p(t)v(t' 2: Q(t)
Q(t)
+4
'lj;(t)
00
+4
[2
l'lj; (t)
a(t)lp(t) ['lj;2(t)
'lj;2()
s ds ] 2
a(s)p(s)
+ 8'lj;(t)
1t
4'lj;2(t)
---:-:-,
-'-:-'-:a(t)p(t)
roo 'lj;2(s) )2
Vt a(s)p(s) ds -
'lj;2(S)
(
a(s)p(s) ds + 16
00
+ 2'lj;2(t) + 'lj;2(t) - 4'lj;2(t)]
4'lj; (t)J
for t 2: to.
= 0
Example 2.3.5. Consider the differential equation (2.3.1) with c ~ 1/4.

Let p(t) = t(lnt)\ t 2: to > 1 where ). < 1 - \1'1- 4c is a constant.
Then, we have
h(t) = _p'(t) =
2p(t)
Let
_~ (~+~)
2
tint
and Q(t) = ().2- 2).+4c)(lnt)>--_2.
ol.(t) = ().2 - 2), + 4c) (I

'f/
4(1 _ ).)
4t
nt
)>---1
Chapter 2
32
Then, 'ljJ'(t) = -Q(t) and
00
'ljJ2(s)
a(s)p(s) ds =
2), + 4c
1
4(1 _ ).)2 'ljJ(t) :::; 4'ljJ(t).
).2 -
It follows from Corollary 2.3.12 that equation (2.3.1) is nonoscillatory if

c:::; 1/4.
Corollary 2.3.13. Let p(t) E C 2([to, 00), JR+) and 'ljJ(t) E Cl([to, 00),
JR). Suppose 'ljJ'(t):::; -Q(t) for t;::: to, and
'ljJl(t) =
00
(IS
'ljJ2(S)
a(s)p(s) exp 2
'ljJ(r)
)
a(r)p(r) dr ds.
If 'ljJl(t) :::; (1/2)'ljJ(t), then equation (2.1.1) is nonoscillatory.

Proof. Let v(t) = -('ljJ(t)
+ 2'ljJl(t))/(a(t)p(t)), t;::: to.
(a(t)p(t)v(t))' ;::: Q(t)
Then, we have
+ 4'ljJ(t)'ljJl(t) + 2'ljJ2(t)
a( t )p( t)
and
Q(t)
:::;
+ a(t)p(t)v2(t) ('ljJ(t) + 2'ljJl (t))2

a(t)p(t)
1
a(t)p(t) ['ljJ2(t)
:::;
(a(t)p(t)v(t))'
_ (4'ljJ(t)'ljJdt) + 2'ljJ2(t))
a(t)p(t)
+ 4'ljJ(t)'ljJl (t) + 4'ljJI(t) -
a(t)~(t) [4 (~'ljJ(t)r _'ljJ2(t)]
4'ljJ(t)'ljJl (t) - 2'ljJ2(t)]
0 for
t;::: to

Corollary 2.3.14. Let the functions p(t), 'ljJ(t)
Corollary 2.3.13. If 'ljJl(t) < 00 for t;::: to, and
and 'ljJl(t) be as in

Proof. Let v(t)
=-
['ljJ(t)
1
+ 'ljJl(t) + 4'ljJ2(t)] /(a(t)p(t)).
(a(t)p(t)v(t))' ;::: Q(t)+ a(t)p(t) [2'ljJ(t)'ljJl(t)
Then, we have
+ 'ljJ2(t) + 8'ljJ(t)'ljJ2(t) + 4'ljJ~(t)]

and for t
Q(t)
33
to,
+ a(t)p(t)v 2(t)
- (a(t)p(t)v(t))'
< a(t)lp(t) {[7/;(t) + 7/;l(t) + 47/;2(t)]2 - [27/;(t)7/;1 (t) + 7/;2(t)

+87/;(t)7/;2(t)
+ 47/;r(t)]
1
[2
a(t)p(t)
7/;1 (t)
2 + 87/;1 (t)7/;2(t) + 167/;2(t)
< a(t)lp(t) [-37/;r(t) + 87/;1(t)
47/;12]
(t)
(~7/;l(t)) + 16 C167/;Ut))]
O.
The following theorem provides a variant of inequality (2.2.16) in Theorem 2.2.3.

exists a function c(t) E C([T, 00), JR) for some T ~ to such that
L6(t)
+ c(t)1 ~
i(3(t)
00
[(3(s) + c(sW dsi

a(s)p(s)
for t
T,
(2.3.6)
where the function (3(t) is defined in (2.2.14).

As an immediate consequence of this result, we have the following corollary.
Corollary 2.3.15. If (3(t) ~ 0 for t ~ to and there exists a function
c(t) E C([T, 00), JR), T ~ to such that
00
[(3(s) + C(S)]2 ds < c(t)

a(s)p(s)
-
for t ~ T

Now, we let C1(t) E C([to,oo),JR) and c(t) = ci(t) - (3(t), where
ci (t) = max{O, c(t)}. Then by Theorem 2.3.3 the following corollary is
immediate.
Corollary 2.3.16. If there exists a function C1 (t) E C([T, 00), JR) for
some T ~ to such that
ci(t)
i{3(t)
00
(Ci(s))2 dsi
a(s)p(s)
for t
T,
34
Chapter 2
Corollary 2.3.17. If
00
rP(s)
1
a(s)p(s) ds :::; 4 1,8(t)l,

Proof. It follows from Corollary 2.3.16 by taking Cl(t) = 21,8(t)l.
Now, let <f>(t) E C1([to, (0), JR) and p(t) E C 2 ([to, (0), JR+) satisfy
<f>'(t) = l/(a(t)p(t)), and let c(t) = (v'k/<l>(t)) - ,8(t), where k > 0 is a
constant. Then, limt--->oo <f>(t) = 00, ,8(t) + c(t) = v'k/<l>(t) and
1 ~(t)+IOO[,8(S)+C(s)j2dsl
a(s)p(s)
t
fJ
I~()
fJ
+ <f>(t)
I.
Thus, we have the following corollary.

Corollary 2.3.18. If there exists a constant k > 0 such that - v'k - k :::;
v'k - k :::; 1/4, then equation (2.1.1) is nonoscillatory.
<f>(t),8(t) :::;
Corollary 2.3.19. If there exists a constant k > 0 such that

Proof. It follows from Corollary 2.3.18 by letting p(t) = 1 and <f>(t)
1 + ft~) l/a(s)ds.
Corollary 2.3.20. Let
C(t) -
."
00
,82(S) exp (IS

,8(7) d7 ) ds,
A
a(s)p(s)
t a(7)p(7)
where A is a constant. If one of the following statements holds:

(i)
A > 1 and ,8(t) > _A_

~(t) - A-I'
(ii) A > 1 and ,8(t) < -A

~(t)
(iii) A = 1 and ,8(t):::;
-~(t),
(iv) A < 1 and _A_ < ,8(t) < _A,

A-I - ~(t) then equation (2.1.1) is nonoscillatory.

Proof. Let c(t) = ().,6(t)[).~(t)
\,6(t)
and
[,6(t)
00
+ ,6(t)])1/2 -
,6(t). Then, we have
+ ).~(t)1
[,6(s) + c(s}F dS\ = 1{3(t)

a(s)p(s)
+ c(t)]2 =
).,6(t)[).~(t)
35
+ ,6(t)]
~ [).~(t)
+ ,6(t)]2,
which implies (2.3.6). Now, it follows from Theorem 2.3.3 that (2.1.1) is
nonoscillatory.
Finally, in this section we state the following results.

Theorem 2.3.4. Let ~(t) and J.L[s, t] be as in Theorem 2.2.4. Then
equation (2.1.1) is nonoscillatory if and only if there exists a function c( t) E
C( [T, (0), JR) for some T ~ to such that
00
e(t) ~
+ e(s)F
a(s)p(s) J.L[S, t]ds.
[~(s)
Theorem 2.3.5.
Let C (t) and J.L* [s, t] be as in Theorem 2.2.5.
Then equation (2.1.1) is nonoscillatory if and only if there exists a function
e*(t) E C([T, (0), lR) for some T ~ to such that
e*(t)
1= [C(~~s;p~:~S)]2
J.L*[S, t]ds.
Taking c(t) = ~(t) and c*(t) = ~*(t) in Theorems 2.3.4 and 2.3.5
respectively, we obtain the following corollary.
Corollary 2.3.21. Equation (2.1.1) is nonoscillatory if
1
4:~(t) ~
00
~2(s)
a(s)p(s) J.L[s, t]ds,
or
1
4:C(t) ~
1=
t
C 2 (s)
a(s)p(s) J.L*[S, t]ds.

In this section, we shall begin with some well-known oscillation criteria
for second order differential equations of the type
(a(t)x'(t))'
where
+ q(t)f(x(t)) =
a(t) E C([to, (0), lR+), q(t) E C([to, (0), lR), to >
COR, JR). In what follows we shall assume that

(i)
0,
Joo ds/a(s) = 00,
and
(2.4.1)
and
f E
Chapter 2
36
(ii) xf(x) > 0 and f'(x)
0 for x#- O.
Theorem 2.4.1. If (i) and (ii) hold, and
00
q(s)ds
(2.4.2)
00.

Proof. Suppose x(t) is a non oscillatory solution of (2.4.1), say, x(t) #- 0
for t ~ tl ~ to. Define wet) = a(t)x'(t)/ f(x(t for t ~ tl' Then for
t ~ h, we have
w'(t)
(2.4.3)
Itt,
Integrating (2.4.3) from h to t, we obtain w(t)::; W(tl) q(s)ds.

Now, assume that x(t) > 0 for t ~ t 1 . The proof in the case x(t) < 0
for t ~ tl is similar and hence omitted. By condition (2.4.2) there exists
t2 ~ tl such that x'(t) < 0 for t ~ t2' Condition (2.4.2) also implies
that there exists a T ~ t2 such that
q(s)ds = 0 and
q(s)ds ~ 0
for t ~ T. Integrating equation (2.4.1) by parts from T to t, we find
J;
It:
a(t)x'(t)
a(T)x'(T) -
q(s)f(x(sds
a(T)x'(T) - f(x(t
f'(x(sx'(s)
(8
q(s)ds
q(U)dU) ds < a(T)x'(T).
Thus, it follows that
x( t) ::; x(T)
t a(1s) ds
+ a(T)x' (T) iT
which is a contradiction.
--> -00
as t
--> 00,
Remark 2.4.1. Theorem 2.4.1 is applicable to the Emden-Fowler equation, namely,

(a(t)x'(t' + q(t)lx(t)I"Ysgn x(t) = 0,
(2.4.4)
where I > 0 is a real constant.
When I = 1 in (2.4.4), we can employ Lemma 2.2.2 to obtain the
following improvement of Theorem 2.4.1.

Theorem 2.4.2.
such that
37
Suppose there exists a function p(t) E C2([t o, 00), lR)
Joo
and
1
--:---:---:-:-ds
a(s)p(s)
Joo Q(s)ds
(2.4.5)
00
(2.4.6)
00,
where Q(t) is defined in (2.2.7), then equation (2.1.1) is oscillatory.

Remark 2.4.2. We can employ Theorem 2.2.1 to extend Theorem 2.4.2
to more general equations of type (2.2.1). The details are left to the reader.
Example 2.4.1. Consider the differential equation (2.3.1). It is easy to
check that Theorem 2.4.1 is not applicable to (2.3.1), in fact, we have
Joo [4~2 + (tl~t)2] ds
<
00,
if c >
However, if we take p( t) = t In t, then
Joo
and
p'(t)
2p(t)
---
h(t)
Joo
p(s) ds
Joo Q(s)ds
-l-ds =
s ns
00
Joo p(s) [q(s) + h 2 (s) oo 4c -1

.
4s In s ds
00
h'(s)] ds
If
c >
Thus all conditions of Theorem 2.4.2 are satisfied if c> 1/4, and hence
(2.3.1) is oscillatory if c > 1/4.
As an immediate consequence of Theorems 2.2.4 and 2.2.5, we have the
following interesting criterion for the oscillation of (2.1.1).
Corollary 2.4.1. Let ';(t) be defined in (2.2.23). If
(J2(s)
(Js
J oo a(s)p(s)
exp 2
or
(3(T)
)
a(T)p(T) dT ds =
e(s)
(2 JS (3(T)a(T)p(T)
+ ';(T) dT) ds
Joo a(s)p(s)
exp
00,
00,
38
Chapter 2
Example 2.4.2.
Consider the Euler differential equation (2.2.11) for
c> 1/4. We let p(t) = ta, where 1 - V4c - 1 ::; 0: < l. Then, we have
(3(t) =
0:
~ ~O:o:+ 4c t a - 1
J= (3:(~~)
and
exp (2
JS ~~~~ dT) ds =
00.
It follows from Corollary 2.4.1 that (2.2.11) is oscillatory if c> 1/4.

Example 2.4.3.
Consider the following differential equation

x /I (t)
where
0:,
-y '" 0,
(0:
siwyt
t17
c) () _
+ -2
t
(2.4.7)
0,
iJ > 0 and c are constants. Let
p(t) = exp (-2
Jt [~-
0:
~~:-YsJ dS),
where A>max{-1/2, (1-2iJ)/2}. Then, p(t)=t- 2A +O(t- 2 >--lT) as

t --+ 00,
pi (t)
A 0: cos -yt
h(t) = - - 2p(t)
t
t er
and as t --+ 00,
We consider the following three cases:

Case (i).
exp
and
If 0 < iJ < 1, we let (1 - 2iJ)/2 < A ::; 1/2. Then as t
--+ 00,
2
(
2-2(7) [1 O( -2>-)]
(2Jt (3(S)d)
p(S) S =
2-y2(2A+2iJ-1)(2-2iJ)
+ t
(JS (3(T)
)
J=(32(S)
p(s)
p(T) dT ds =
0:
exp
exp
00.
It follows from Corollary 2.4.1 that (2.4.7) is oscillatory if 0

Case (ii). If iJ
=1
and c> (1/4) -
(0: 2/2-y2),
we let
<c<
l.
39
Then,
(3( t) -
2')'2(,),2 +,), + c) +
2')'2(1 + 2,),)
2 -2>'-1
0 ( -2>.-2)
t
exp (2 /t (3(s) dS) = t(2,),2(>.2+>.+c)+a 2)/'"'?(1+2>') [1 + 0
p(s)
and
00
/
(/8 (3(T)
)
p(T) dT ds =
(32(s)
p(s) exp 2
as
-> 00,
(C 2>'-1)] as t -> 00
00.
It follows from Corollary 2.4.1 that (2.4.7) is oscillatory if u = 1 and

c> (1/4) - (0 2/2')'2).
u>
Case (iii). If
1 and c> 1/4, we let
~2
< ,), <

min {
-
~2'
~+
- 2
~}
4'
c-
Then,
(3(t)
),2 +), + c C 2>'-1 + 0

1 + 2),
(t- 2>.-2)
exp(2/t(3(S)dS) = t 2(>.2+>'+C)/(1+2>') [1+

p(s)
and
00
/
as
-> 00,
o (t- 2>.-1)]
(/8 (3(T)
)
p(T) dT ds =
(32(S)
p(S) exp 2
t->oo
as
00.
It follows from Corollary 2.4.1 that (2.4.7) is oscillatory if u

c> (1/4).
>
1 and
Under an additional condition on q(t),
(2.4.2) can be replaced by
lim sup rt q(s)ds =
T 2': to.
t-Hx)
iT
00
for all large
(2.4.8)
In fact, we have the following result.

Theorem 2.4.3. Assume that (i) and (ii) are satisfied. If
lim inf rt q( s )ds 2': 0
t---+oo
iT
for all large
T 2': to
(2.4.9)
and condition (2.4.8) holds, then equation (2.4.1) is oscillatory.
Proof. Let x(t) be a nonoscillatory solution of equation (2.4.1), say,

x(t) > 0 for t 2': t1 2': to. As in Theorem 2.4.1, we obtain (2.4.3). Next,
we consider the following three cases:
40
Chapter 2
Suppose x'(t) changes signs. Then there exists a sequence

limn--->oo Tn = 00 such that x'(Tn) = O. Choose N
large enough so that (2.4.9) holds. We then have
Case 1.
{Tn};:"=l with
a(t)x'(t) < a(TN )x' (TN) - it q()d

s s,
f(x(t)) f(x(TN))
TN
so
a(t)x'(t)
hmsup
t--->oo f(x(t))
< a(TN)x'(TN).
+ hmsup (- i t q(s)ds )
-
f(x(TN))
t--->oo
TN
< 0,
which contradicts the fact that x' (t) oscillates.

Case 2. Suppose x'(t) > 0 for all t
a(t)x'(t)
f(x(t))
--7-:~7-:-'-
a(t2)x'(t2)
f(X(t2))
t2
tl, then from (2.4.3), we have
+ i t q (s )ds < 0 ,
t2
and from (2.4.8), we obtain lim inft--->oo a(t)x'(t)/ f(x(t)) = -00, which is
a contradiction.
Case 3. Suppose x'(t) < 0 for t ~ t3 ~ to. Condition (2.4.9) implies
that for any t4 ~ t3, there exists a T ~ t4 such that
q(s)ds ~ 0 for
all t ~ T. Choosing T ~ t3 as indicated, and then integrating (2.4.1), we
obtain
J;
a(t)x'(t) ::; a(T)x'(T) - [
q(s)f(x(s))ds
= a(T)x'(T) - f(x(t)) itq(s)ds
+ It f'(x(s))x'(s) is q(u)duds
::; a(T)x' (T).
J;
Thus, x(t)::; x(T) + a(T)x'(T)

l/a(s)ds
this contradicts the fact that x(t) > 0 for t
-->
tl.
00
as t
--> 00.
But,

In this section we shall employ averaging techniques to study the
oscillatory behavior of equations (2.1.1) and (2.2.1), so that the criteria
we shall present will involve the average behavior of the integral of the
alternating coefficients. The motivation of this criteria comes from the
disconjugate property of (2.1.1) on [to, 00), to ~ O.
Theorem 2.5.1. If
/00 (1: a(U)dU)
-1
ds
00
(2.5.1)

and
t->oo litis
to to q(u)duds
lim -
(2.5.2)
00
41

x(t) > 0 for t ~ to> O. Define w(t) = a(t)x'(t)/x(t) for t ~ to. Then,
we have
1
w'(t) = - q(t) - a(t) w 2(t) for t ~ to
(2.5.3)
Integrating (2.5.3) twice from to to t, we obtain
it,t
w(s)ds
it is -(2-) duds
to to
w (u)
a u
(k - ! it is q(U)dUdS) ,
t
to
where k is a constant. By (2.5.2), there exists a tl
i tw(s)ds + itto is w2u duds <

to
Thus for t
a u
to
to
for
to such that
t
~ tl.
tl it follows that
F(t)
itoti8 -(-) duds < - it w(s)ds.

w 2 (u)
to
a u
to
Since F(t) is nonnegative, we get
F2(t) <
(1:
(2.5.4)
W(S)dS) 2
By Schwartz's inequality, we have
It is easy to check that there exists a t2

t ~ t2, and hence
(1~ a(S)ds)
-1
tl such that F(t) > 0 for
< F- 2(t)F'(t)
for
~ t2.
Now, integrating (2.5.5) from t2 to t, we get
1: (1:
a(U)du) -1 ds
<
00,
(2.5.5)
Chapter 2
42
which contradicts condition (2.5.1).
From Lemma 2.2.2 and Theorem 2.5.1 the following corollary is immediate.
Corollary 2.5.1. If there exists a function p(t) E C 2 ([to, 00), 1R+) such
that
j= (1: a(U)p(U)dU)
litis
and
lim -
t .....
=t
to
to
-1
ds
00
= 00,
Q(u)duds
(2.5.6)
(2.5.7)

Similarly, from Theorems 2.2.1 and 2.5.1 the following corollary is immediate.
Corollary 2.5.2. If there exists a function p(t) E C 2 ([to,00),1R+) such
that condition (2.5.6) holds and
litis
lim t .....
=t
to
til
Q*(u)dud.c;
00,
where Q*(t) is defined in (2.2.3), then equation (2.2.1) is oscillatory.

1
)'
( "tx'(t)
+ t12X'(t) + t~x(t) =
0,
t ~ 1
(2.5.8)
where c is any positive constant. Let p(t) = t. Then, a(t)p'(t)-p(t)p(t) =

o and Q*(t) = p(t)q(t) = cit. Thus, all conditions of Corollary 2.5.2 are
satisfied, and hence (2.5.8) is oscillatory for all c> O.
Taking p(t) = t in Corollary 2.5.1, it follows that the undamped
equation associated with (2.5.8), namely, the equation
(~X'(t)' + t~ x(t)
= 0
(2.5.9)
is oscillatory, and hence the damping term in (2.5.8) preserves oscillations.

Remark 2.5.1. In Theorem 2.5.1 and Corollaries 2.5.1 and 2.5.2 if we let
(2.5.10)
Oscillation and non oscillation of linear ordinary diHerential equations
43
where al is a positive constant, then (2.5.1) is superfluous and condition

(2.5.6) becomes
(2.5.11)
Next, we shall show that (2.5.2) can be replaced by two weaker conditions
lim inf
t->oo
and
1
lim sup -T
rt q(s )ds
iT
> - A
ftfs q( u )duds =
T
t-~oo
(2.5.12)
(2.5.13)
00,
for all large T. In (2.5.12), A is a positive constant.

Theorem 2.5.2. If conditions (2.5.10), (2.5.12) and (2.5.13) hold, then
equation (2.l.1) is oscillatory.
Let x(t) be a nonoscillatory solution of equation (2.l.1), say,

x(t) > 0 for t::.> T ::.> to with T::.> l. From (2.l.1), we have
Proof.
a(t)x'(t)
x(t)
rt () (x'(S))2
t ()d =
d
x(s)
s+ iT q s S
+ iT a s
a(T)x'(T)
x(T)
c. (2.5.14)
We need to consider the following three cases:

Case l. Suppose x'(t) has arbitrarily large zeros, i.e., there is a sequence
{tn}, tn - t 00 as n - t 00 such that x'(t n ) = O. Evaluating (2.5.14)
along this sequence, we obtain
(t)
(a(t)) 1/2 X'
x(t) E
By Schwartz's inequality, we have
([
a(;~)(s) dS) ~
2
alt
It
r2 (T
)
L
,00.
(:(~}
a(s)
for some positive constant k. This implies

and so from (2.5.14), we get
-k +
~
t
rt t
iT iT
q(u)duds
(2.5.15)
ds
k2t2
J;a(s)x'(s)/x(s)ds::.> -kt
c(t - T) < c,
t

Case 2. If x'(t) > 0 for t::.> Tl ::.> T,
(2.5.15), and we can proceed as above.
then from (2.5.14) we again have
Chapter 2
44
Case 3. Suppose there exists tl :::: T such that x'(t) < 0 for t:::: tl.
From (2.5.14) and (2.5.12), we have
_ a(t)x'(t) > _ (
')
_
c + /\
x (t )
it ()(X'(S))2
t,
a s
( )
X S
ds.
(2.5.16)
If the above integral converges, then we can proceed as in the previous

cases. If the integral diverges, then there exists a t2:::: tl such that
t2
t,
(X'(S))
a(s)
x(s)
1+
ds
+ A.
Multiplying (2.5.16) by
-(x'(t)/x(t))
-(c + A)
+ Jtrt, a(s) (x'(s)/x(s)) 2 ds
and integrating from t2 to t, we obtain

In
[-(C + A) + l>(S) (:(~i dS]
> In :((:/'
which by (2.5.16) implies that -(a(t)x'(t)/x(t)):::: (X(t2)/X(t)) for t:::: t 2 .

Hence, x'(t) :<::: -x(t2)/al for t :::: t2, and now final integration shows that
x(t) must become eventually negative. This contradicts the assumption
that x(t) > 0 for t:::: T.
Remark 2.5.2. In the proof of Theorem 2.5.2 the three cases considered
depending on the sign of x' (t) can be replaced by only two cases depending
on the integral
ft
(x'(s)) 2
x(s)
ds
JT a(s)
to be either convergent or divergent. This technique will be used often in

the proofs of later results.
Now, we shall show that (2.5.7) can be replaced by two weaker conditions
11 1
liminf t-->oo
to
to
Q(u)duds > -
and that limit in (2.5.7) does not exist, i.e.,

I~~~f
1
-
j.t j.s Q(u)duds

to
to
< lim sup -1

t-->oo
00
1t 1s
to
to
Q(u)duds <
00.
45
Theorem 2.5.3. If there exists a function pet) E C2([to, (0), JR+) such
that
lit
lim 2"
t--->oo t
to
(2.5.17)
a(s)p(s)ds = 0
and
-00
<
lil~~f~iti8 Q(u)duds
t
to
to
<
limsuP~iti8 Q(u)duds
t--->oo
to
:::::
00,
to
(2.5.18)
Remark 2.5.3. In Theorem 2.5.3 if aCt) = pet) = 1, then (2.5.17) is
disregarded and (2.5.18) reduces to the well-known condition, namely,
-00
<
lit~~f ~t
it i8
to
to
q( u )duds < lim sup ~

t--->oo
it i 8
to
to
q( u )duds :::::
00.
(2.5.19)
The proof of Theorem 2.5.3 is contained in a more general result which
uses weighted-integral condition. For this, we assume that the function
pet) E C 2 ([to, (0), JR+) is given, h(t) = -p'(t)/(2p(t)) and Q(t) =
pet) [q(t) + a(t)h 2(t) - (a(t)h(t))'] .
Let n(p) be the set of all nonnegative locally integrable functions get)
on [to, (0) satisfying the condition
li~~p
(j 9(S)dS)
t
l-k
[Gk(oo) - Gk(t)] > 0 for some
k E [0,1),
(2.5.20)
where
Gk(t) =
If Gk(oo) =
00,
g(s)
(f g(u)du)k
(f a(u)p(u)g2(u)du) ds.
(2.5.21)
then 9 E n(p).
Let no (p) be the set of all nonnegative locally integrable functions

get) on [to,oo) satisfying
lim
t--->oo
Joo
a(s)p(s)g2(~)ds = O.
(t g(s)ds)
(2.5.22)
In order that either (2.5.20) or (2.5.22) is satisfied by a nonnegative

locally integrable function g( t), it is necessary that
joo
g(s)ds =
00.
(2.5.23)
Chapter 2
46
Members of the classes D(p) and Do(p) are called weighted functions with respect to p(t). It follows from (2.5.20) and (2.5.22) that
Do(p) c D(p).
For 9 E D = D(p), we define
_ I: g(u) IsuQ(T)dTdu
Is g(u)du
Ag(s, t) -
(2.5.24)
If (2.1.1) is nonoscillatory, then as in Theorem 2.2.3, equation (2.2.18)

holds for t::=: T::=: to, where v(t) = a(t)p(t)w'(t)/w(t), t::=: T i.e.,
I
v (t)
v 2 (t)
+ Q(t) + a(t)p(t) =
(2.5.25)
t::=: T.
for
Clearly, an equivalent integral equation of (2.5.25) is
rt a(u)p(u)
v (u)
rt
du - is Q(u)du.
2
is
v(t) = v(s) -
(2.5.26)
Lemma 2.5.1. Let v(t) be a solution of equation (2.5.25). If there exists

a function g(t) E D satisfying
liminf A g (., t) > then
00
/
(2.5.27)
00,
t-->oo
V2(S)
a(s)p(s) ds <
00.
v2(s)
a(s)p(s) ds
00.
(2.5.28)
Proof. Assume that

00
/
(2.5.29)
Multiplying (2.5.26) by g(t) and integrating it from

we obtain
tg (S)V(S)dS = V(T)
it -it is
i is
g(s)ds
g(s)
= [V(T) - Ag(T, t)]
g(s)
to t, t ::=:
::=: T,
Q(u)duds
v2 (u)
a(u)p(u) duds
(i
g(s)ds -
it is
T
g(s)
v 2 (u)
a(u)p(u) duds.
(2.5.30)
Oscillation and nonoscil1ation of linear ordinary differential equations
47
From equation (2.5.26), we have
V(T) = v(T) Since g(t) E D,
r Q(s)ds - Jr
JT
v 2 (s)
a(s)p(s) ds.
(2.5.23) holds, and hence
r Q(s)ds _ I; g(s)/; Q(u)duds
I~ g(s)ds Ag(T, t) _
IT g(s)ds
IT g(s)ds
JT
+ 0(1)
Ag(T, t) - / : Q(s)ds
as
--> 00.
v 2 (s)
V(T) - Ag(T, t) = v(T) - Ag(T, t) - JT a(s)p(s) ds
+ 0(1)
as
Since g(t) E D, there exists a positive number ,\ > 0 such that
--> 00.
(2.5.31 )
(2.5.32)
where k is as in (2.5.21). It follows from (2.5.27), (2.5.29) and (2.5.31)
that there exist two numbers tl and t2 with t2:;': tl :;,: T such that
v(tl)-Ag(tl,t) :::: -,\

Let y(t)
Ittl
forall
t:;,:t2.
(2.5.33)
g(s)v(s)ds. Then Schwartz's inequality gives
tl
V2 (u) d
u >
a(u)p(u)
-
y2 (s )
Itls a(u)p(u)g2(u)du .
It follows from (2.5.30) and (2.5.33) that
y(t) < _,\

-
jt g(s)ds -jt It, a(u)p(u)g2(u)du

g(s)y2(s)
ds
s
it
tl
- F(t). (2.5.34)
Thus, we find
F'(t)
,\g(t)
g(t)y2(t)
a(s)p(s)g2(s)ds
(2.5.35)
< F(t) < ly(t)l.
(2.5.36)
Itl
and
o :::: ,\
jt g(s)ds
tl
Chapter 2
48
Now, for t 2: tl from (2.5.34) - (2.5.36), we obtain
)-1
i t ) k (it
F k- 2(t)F'(t) 2: Fk(t)F'(t)
y2(t)
2: )..kg(t) ( t,g(s)ds
tla(s)p(s)l(s)ds
Integrating the above inequality from t 2: t2 to z and letting z -> 00,

and noting that k E [0,1), we get Fk-l(t) 2: (1- k) ..k[Gk(oo) - Gk(t)].
Finally, it follows from (2.5.36) that
2: (1- k)
(1:
9(S)dS) l-k [Gk(oo) - Gk(t)],
which contradicts (2.5.32). This completes the proof.
Lemma 2.5.2. If equation (2.5.25) has a solution v(t) satisfying (2.5.28),

then for any g(t) E no = no(g), limt-->oo Ag(, .t) exists, and
hm Ag(T,t) = V(T) t-->oo
v2 (s)
( ) ( )ds.
T asp s
00
(2.5.37)
Proof. As in Lemma 2.5.1, (2.5.30) holds and this implies that

_
A(T, t) = V(T)
JTt 9 ()
s V (s )ds
ItT 9 (s )JST a(u)p(u)

v 2(u) d d
U S
I: g(s)ds
(2.5.38)
I: g(s)ds
Since 9 E no, (2.5.23) holds. Thus, it follows that
. I: g(s) I: a(:)~~)duds
11m
t
< 00.
t-->oo
IT g( s )ds
Next, by Schwartz's inequality, we have
. II: g(s)v(s)dsl
o <
hm
t-->oo
<
hm
t-->oo
-'-----t:;:-----'-
. u:
IT g( s )ds
a(s)p(s)g2(s)ds) 1/2
t
U: a(s2)~(s)ds)I/2
ITg(s)ds
O.
Hence, from (2.5.38), limt-->oo Ag(T, t) exists and (2.5.37) holds.
From Lemmas 2.5.1 and 2.5.2 the following theorem is immediate.

Theorem 2.5.4. Equation (2.1.1) is oscillatory if there exists a function
g(t) E no such that

-00
< liminfAg(.,t) < limsupAg(.,t) <

t~oo
t~oo
00.
(2.5.39)
49
Further, following Lemma 2.5.1 we can easily prove the following result.
Theorem 2.5.5. Assume that there exist two functions p(t) E C 2([to, 00),
JR+) and g(t) E n such that Iimt-+oo Gk(t) = 00 for some k E [0,1)
and limt-+oo A g (., t) = 00, then equation (2.1.1) is oscillatory.
Example 2.5.2. Consider (2.3.1) with c> 0. If p(t)
h(t) =
- 2t - 2tlnt
4c - 1
Q(t) = 4tlnt
and
= tInt, then
t 2: T> 1.
for
Let g(t) = l/(tlnt), t 2: T. Then, we have

lim Gk(t)
t-+oo
= t-+oo
lim
t [Inlns -lnlnT]k-l
I
ds
T
S ns
= t-+oo
lim [lnlnt -lnlnT] =
00
and
lim Ag(T, t)
t-+oo
1) 1t 1
1)
4C -lim ( t-+oo
4
4C lim ( - t-+co
4
--->
00
as
--ds
sIn s
[Inlnt -lnlnT]
---> 00
when
> 4"
where T> 1 is large enough. Hence, by Theorem 2.5.5 equation (2.3.1)

is oscillatory if c> 1/4.
Remark 2.5.4. Theorem 2.5.3 is an application of Theorem 2.5.4. In
fact, if we let g(t) = 1, then from (2.5.17) and (2.5.18) it follows that
g(t) E no and Iimt-+oo A g (., t) does not exist. Thus, by Theorem 2.5.3,
equation (2.1.1) is oscillatory.
As further applications of Theorem 2.5.4, we have the following results:

Corollary 2.5.3. Let k(t) E C1([to, 00), JR+)
l/(a(t)p(t)) and limt-+()() k(t) = 00. If
-00
"
1
< hm
lilf k( t )
t-+()()
It IS
to
to
< lim sup - 1
t-+oo k(t)
be such that
Q(u)
a (u ) P(u ) duds
1t IS
to
to
Q(u) duds
a(u)p(u)
k'(t)
(2.5.40)
:s:
00,
Chapter 2
50
Let get) = l/(a(t)p(t)) = k'(t). It follows from (2.5.40) that

and limt->oo A g (., t) does not exist. Thus, by Theorem 2.5.4,
Proof.
get)
no
Corollary 2.5.4. Suppose there exist two nonnegative bounded functions gl(t) and g2(t) on [T,oo), T 2': to satisfying Joo gl(s)ds =
00 = Joo g2(s)ds, and a(t)p(t)gl (t) and a(t)p(t)g2(t) are bounded. If
limt-->oo A g, (T, t) < limt->oo Ag2 (T, t), then equation (2.1.1) is oscillatory.
Proof. Let al and a2 be two numbers satisfying limt-->oo A g , (T, t) <
al < a2 < limt--> 00 A g2 (T,t). Let ret) = g2(t) for T::; t < tl, where tl
is such that Ar(T, it) > a2 and I~' r(s)ds 2': 1. Let ret) = gl(t) for
it ::; t < t2, where t2 is such that Ar(T, t 2) ::; al and I~2 r(s)ds > 2.
This is possible because
J~2 res) J; Q(u)duds

J~2 r(s)ds
J~'[g2(S) - gl(S)] I;Q(u)duds

I~' g2(s)ds
+ Itt,2 gl(s)ds
( J~29l(S)J;Q(U)dUdS)
J~2 gl(s)ds
A g , (T, t2)[1
+ 0(1)] + 0(1)
(
as
J~'
t
J~2gl(s)ds
g2(s)ds + Itt,2 gl(s)ds

--t
00.
Continuing in this manner, we obtain a nonnegative, nonintegrable and

bounded function r(t) defined on [T, (0) such that lim inft-->oo Ar(T, t) ::;
al < a2 ::; lim SUPt-> 00 Ar(T, t) and
hm
t->oo
J;a(s)p(s)r2(s)ds
t
<ITr(s)ds)2
= o.
This implies that r(t) E no and limt->oo Ar(T, t) does not exist. Thus,
by Theorem 2.5.4, equation (2.1.1) is oscillatory.
Remark 2.5.5. It follows from the results presented above and Theorem
2.2.1 that in the hypotheses of each of the Theorems 2.5.2 - 2.5.5 the
function Q(t) can be replaced by Q*(t) defined in (2.2.3).
(t 2x' (t))' - 3tx' (t)
+ ~ev't(2 + cos t - 2t sin t)f(x(t))
0,
7r
t>to>-2
(2.5.41)
51
where the function f(x) is one of the following:
f(x) = mx, x E lR, m is a positive constant,

f(x) = Ixl">' + mx, x E lR, m and 'Yare positive constants,
(i)
(ii)
(iii) f(x) = x In2 (JL + Ix!), x E lR, JL> 1 is a constant,

(iv) f(x) = x exp(Ax), x E lR, A>
is a constant,
(v)
f(x) = sinh x, x E lR.
We take p(t) = 1/t 3 and g(t) = t, t ;:::: 7r/2. Then, we have h*(t) =
(p(t)p(t) - a(t)p'(t))/(2k a(t)p(t)) = 0, and hence Q*(t) = p(t)q(t)
(2 + cost - 2tsint)/(2v'i). Now, since
p(s)q(s)ds
i [0
i
~/2
to
sin s +
d(0(2
1;;:;(2 + cos S)] ds
2yS
+ coss))
~/2
Vi(2
and
.t
ito g( s )ds
;::::
lt (l
g(s)
to
t -
+ cost) -
2(7r/2)1/2 >
Vi -
2(7r/2)1/2
p(u)q(U)dU) ds
to
1
(/2)2
7r
it;;:;
(sys-2(7r/2)
1/2
~/2
s)ds
00
as
t~oo
all conditions of Theorem 2.5.5 with the function Q replaced by Q* are

satisfied and hence equation (2.5.41) is oscillatory.
o ds/a(s) =
We note that in Theorems 2.5.2 - 2.5.5 condition
not required. We also note that condition (2.5.2) is equivalent to
11t
lim -
t-->oo
to
(t - s)q(s)ds =
which in turn implies that for any integer n

lim -1
t-->oo
tn
lt
to
00,
00
is
(2.5.42)
>1
(t - s)nq(s)ds =
00.
(2.5.43)
It is to be further noted that (d. Remark 2.5.3) in (2.5.42) the limit cannot
be replaced by an upper bound.
Now, we state the following oscillation criterion for the linear differential
equation
xl/(t) + q(t)x(t) = 0,
(2.5.44)
Chapter 2
52
where q(t) E C([to, (0), lR).
Theorem 2.5.6. If for some integer n
> 1,
limsup -1 i t (t - s)nq(s)ds
tn
t-+oo
(2.5.45)
00,
to

Further, in the case when condition (2.5.45) is violated, we have
Theorem 2.5.7. Let
limsup
t-+oo
~n i \ t t
to
s)nq(s)ds <
00
n>
for some
l.
If there exists a function n(t) E C([to, (0), lR) such that
liminf
t-+oo
~
t(t-s)nq(s)ds ~
t n iT
n(T)
T~to
forevery
and ft,:(n+(s))2ds = 00, where n+(t) = max{n(t),O}, t ~ to then

An extension of Theorem 2.5.6 is as follows:
Theorem 2.5.8. Suppose H: V
continuous function such that
H(t, t)
0 for
to,
= {(t, s)
t ~ s ~ to}
H(t, s) > 0 for
t >s
-->
to
lR is a
(2.5.46)
and has a continuous and nonnegative partial derivative on V with respect

to the second variable. Let h: V --> lR be a continuous function with
- 8s H(t, s) = h(t, s)JH(t, s)
for all
(t, s)
(2.5.47)
V.
If
lim sup H( 1
t-+oo
t, to
) i t [H(t, s)q(s) - -41 h2(t, s)] ds

to
00,

Remark 2.5.6. Oscillation criteria stated in Theorems 2.5.6 - 2.5.8 for
(2.5.44) are of very limited applications. In fact, these fail to deduce the
oscillatory character of Euler's equation (2.2.11).
These criteria can be extended rather easily to equation (2.1.1) by

applying Lemma 2.2.2. In fact, we only need to assume that p(t) E
53
C1([to, (0), IR+), and replace the functions a(t) and q(t) by a(t)p(t)
and Q(t) = p(t) [q(t) + a(t)h 2(t) - (a(t)h(t))'] respectively, where h(t) =
-p'(t)/(2p(t)).
Further extensions and improvements of these criteria to more general
equations of type (2.2.1) can be obtained by applying Theorem 2.2.1. For
this, we let p(t) E C1([to,00),IR+), and replace the functions a(t) and
q(t) by a(t)p(t) and kQ*(t) where Q*(t) = p(t)[q(t) + a(t)h 2*(t) (a(t)h*(t))'], h*(t) = (p(t)p(t) - a(t)p'(t))/(2ka(t)p(t)) and the constant
k is as in (i2), i.e., f'(x) 2 k > 0 for x # o.
In the following result we extend Theorem 2.5.8 to equation (2.1.1).
Theorem 2.5.9. Let the functions Hand h be as in Theorem 2.5.8
such that (2.5.46) and (2.5.47) hold. If
lim sup
t->oo
(1
H t, to
It
to
[H(t, s)q(s) -
~h2(t, s)a(s)]
ds
00,
(2.5.48)

Proof. Let x(t) be a nonoscillatory solution of (2.1.1), and let To 2 to
be such that x(t) # 0 for t 2 To. Define v(t) = a(t)x'(t)/x(t) for t 2 To.
Then from equation (2.1.1) it follows that q(t) = -v'(t) - (v 2(t)/a(t)) for
t 2 To Thus, for every t, T with t 2 T 2 To, we obtain
rt
iT H(t, s)q(s)ds
rt
rt
v2 (s)
iT H(t, s)v'(s)ds - iT H(t, s) a(s) ds

rt [ a
v 2(S)]
H(t, T)v(T) - iT - as H(t, s)v(s) + H(t, s) a(s) ds
= -
H(t, T)v(T) - [
H(t, T)v(T) -
11t
+-
iTt
[h(t, s)JH(t, s)
[(H(t
a(;)s
+ H(t, s) :(~)]
ds
))1/2 v(s) + "21 Va(;)h(t, s) ]2 ds
h2(t, s)a(s)ds.
Hence, for every t 2 To, we have
J~ [H(t, s)q(s)- ~h2(t, s)a(s)] ds::;
H(t, To)v(To)
::; H(t, To)lv(To)1 ::; H(t, to)lv(To)l.
Chapter 2
54
Therefore, for all t
To it follows that
1(~ [H(t, s)q(s) - ~h2(t, s)a(s)] ds

= rTofH(t, s)q(s) Jto
::;
in
~h2(t, s)a(s)l ds +
J
rt fH(t, s)q(s) -
JTJ
H(t, s)lq(s)lds + H(t, to)lv(To)1 ::; H(t, to)
[in
~h2(t, s)a(s)l ds
J
Iq(s)lds + Iv(To)1 .
This gives
lim sup H( 1
t, to
t-->oo
it [H(t, s)q(s)- "4h2(t,

1
s)a(s) ] ds::; Iv(To)l+ iTO Iq(s)lds,
to
to
which contradicts (2.5.48).
Now, in view of Remark 2.5.6 we can state the following more general
results.
Theorem 2.5.10. Let H, h be as in Theorem 2.5.8 and let there exist
a function p(t) E C1([to,00),lFt+) such that
lim sup
t-->oo
(1
H t, to Jto
[H(t,S)Q(S) -
~h2(t,s)a(s)p(s)]
ds
4
= 00, (2.5.49)

Theorem 2.5.11. Let H, h be as in Theorem 2.5.8 and let there exist
a function p(t) E C 1([to,00),1R.+) such that
lim sup H( 1
t-->oo
t, to Jto
[H(t, s)Q*(s) - 41kh2(t, s)a(s)p(s)] ds = 00,
(2.5.50)
where Q*(t) and k are defined in Remark 2.5.6, then equation (2.2.1) is
oscillatory.
hto
Let R(t) =
dsja(s). If there exists a function
p(t) E C1([to, 00), IR+) such that for some n> 1,
Corollary 2.5.5.
li~~p R';(t) 1:(R(t)-R(s))n-2 [(R(t)-R(S))n-2Q*(s)- ~~ :~:~] ds =
00,
(2.5.51 )
55
Let H(t, s) = (R(t) - R(s))n for t ~ s ~ to. Then, h(t, s) =

(n/a(s))(R(t) - R(s))(n-2)!2 for t > s ~ to, and this implies
Proof.
lim sup H( 1
t, to
t-+CXJ
= lim sup Rn1()
t-+oo
ltDt
[H(t,S)Q*(s) - 41ka(s)P(S)h2(t,s)] ds
rt [(R(t)-R(s))nQ*(s)- 4nk~ ap((s))s (R(t)-R(S))n-2] ds
ltD
00.
Now, it follows from Theorem 2.5.11 that (2.2.1) is oscillatory.

If p(t)
ltD
1, then we have
a(s)p(s)h2(t,s)ds
~(R(t) -
n- 1
R(to))n-l
for
~ to
and hence the following corollary is immediate.
Corollary 2.5.6. If
1 )
lim sup Rn(
t
t-+oo
it
to
(R(t) - R(s)tQ*(s)ds
for some n> 1, (2.5.52)
00
Remark 2.5.7. Condition (2.5.52) for equation (2.1.1) reduces to

lim sup R:()
t-+CXJ
t (R(t) -
ltD
R(s))nq(s)ds =
Let Rl(t)
Corollary 2.5.7.
n> 1,
00
= ftCXJ ds/a(s) <
for some
00,
t ~ to.
n> 1.
If for some
Proof. Let p(t) = Rl(t) for t ~ to and H(t,s) = (In(Rl(s)/Rl(t))t

for t ~ s ~ to. Then, we have h(t) = -p'(t)/(2p(t)) = 1/(2a(t)R1 (t)),
Q(t) = p(t) [q(t)
+ a(t)h 2(t) -
(a(t)h(t))']
q(t)Rdt) - 4a(t)Rl(t)
and
h(t,s)
n
(Rl(S))(n-2)!2
a(s)Rl(S) In Rl(t)
for
> s ~ to
Chapter 2
56
Thus, for t 2: to it follows that
a(s)p(s)h2(t,s)ds
17,
to
to
2 (
( )n-2
1 R IS
1
d
n Rl(t)
a(s)Rl(S) s
2
(
R (
n-l <
_17,_
ln~
n-1
R 1 (t)
00
It is easy to check that (2.5.49) implies condition (2.5.53), and hence equation (2.1.1) is oscillatory.
I
Next, we let H(t, s) = (t - s)n for t 2: s 2: to, where n > 1 is a
constant. Then from Theorem 2.5.11 the following corollary is immediate.
Let n > 1 be a constant and there exist a function
Corollary 2.5.8.
p(t) E C 1 ([to, 00), lR+) such that

lim sup n1
t~oo
it
to
7 , 2 ] ds
(t - s)n-2 (t - s)2Q*(S) - 1
-ka(s)P(s)
00,

(t 2x'(t' - tx'(t)
+ cf(x(t
0,
t 2: to 2: 1
(2.5.54)
where c > 0 is a constant and f(x) is as in Example 2.5.3 with f' (x) 2:
k > 0 for x ~ O. Let p(t) = lit, t 2: 1 and n> 1 a number. Then, we
find a(t)p'(t)-p(t)p(t) = 0, and hence h*(t) = 0 and Q*(t) = cit, t 2: 1.
Now,
limsup
t~oo
it [(t to
s)nQ*(s) - nk2a(s)p(s)(t - s)n-2] ds

4
li~~p t~
it
[(t - s)n
(~) - ~~(t -
s)n- 2s] ds.
Since n> 1 it follows from Hardy et. al. [10] that (t - s)n 2: t n - nst n - 1
for t 2: s 2: 1. Thus,
lim sup n
C
t~oo
jt -(t - s)nds
1
tIS
2: clim sup t~oo t n
jt tn _ 17,st
s
n(t -1)]
n- 1
climsup [lnt _
t~oo
ds
=
00.
Also, we have
n2
4(n -1) (t _l)n-l
+ 4(n -1) (t -It,
57
and hence
lim sup ~ {t [~(t _ st _ n 2 s(t _ s)n-2] ds
t-HOC
tn ) 1 s
4k
=climsup [lnt_ n (t-1)]_
t
t~oo
(n )k =00
4 n-1
if c>O.
Now, it follows from Corollary 2.5.8 that (2.5.54) is oscillatory if c > O.

(tx'(t))' +x'(t)
+ "if(x(t)) =
0,
2: 1
(2.5.55)
where c> 0 is a constant and f(x) is as in Example 2.5.4. Let p(t) = 1.

Then, h*(t) = 0 and Q*(t) = cit, t 2: 1. Now, proceeding exactly as
in Example 2.5.4, we find that (2.5.55) is oscillatory by Corollary 2.5.8 if
c> O.
In Theorems 2.5.6- 2.5.11 no assumption has been made on Joo dsla(s).
Therefore, the conclusions of these criteria hold in either of the following
two cases:
(I)
Joo dsla(s)
= 00,
(II)
Joo dsla(s)ds < 00.
For illustration see Examples 2.5.4 and 2.5.5.

We note that if (II) is satisfied, then the condition (2.5.51) is equivalent
to the following
lim sup
t-.oo
for some
It
to
(R(t) - R(s))n-2 [(R(t) - R(s))2Q*(t) - n42 PS))] ds

a s
n>
00
1.
Also, it is easy to show that the condition
liminfRl(t)It Q(s)ds >
t--->oo
to
~4
(2.5.56)
implies (2.5.53) with n = 2. Thus, we have the following corollary.

Corollary 2.5.9. Let condition (2.5.53) of Corollary 2.5.7 be replaced by
(2.5.56), then equation (2.1.1) is oscillatory.
We are now in the position to prove the following oscillation criteria
which extend and improve Theorem 2.5.7.
Cbapter 2
58
Theorem 2.5.12. Let Hand h be as in Theorem 2.5.8 and conditions

(2.5.46) and (2.5.47) hold, and let
. { . . H(t,s)}
o < s~to
mf hm
mf H( t, to )
t--->oo
(2.5.57)
00.
Suppose there exist two functions p E C 1 ([to: oo),Rn and DE C([to, (0),
R) such that
it
i
lim sup H( 1 )
t--->oo
t, to
to
oo
to
and for every T

lim sup
t--->oo
a(s)p(s)h2(t, s)ds <

(D+(S))2
--,---;-...,-'--:-,-:-ds
a(s)p(s)
(2.5.58)
00,
(2.5.59)
00
to,
(1 )
H t, T
s)a(s)p(s)] ds ~
iTrt [H(t, s)Q(s) - ~h2(t,
4
where Q is defined in (2.2.7) and D+(t)

(2.1.1) is oscillatory.
= max{D(t), O},
D(T),
(2.5.60)
then equation
Proof. Assume to the contrary that equation (2.1.1) is nonoscillatory.

Let the function v(t) be as in Theorem 2.2.3. Then, we have
I
v (t)
v 2 (t)
+ Q(t) + a(t)p(t)
= 0
t ~ To ~ to
for
Multiplying both sides of the above equation by H(t, s) and proceeding

as in Theorem 2.5.9, we get
H(:,T)
it
= v(T) - H(t, T)
for t
~a(s)p(S)h2(t,S)] ds
[H(t,S)Q(s) -
H t
(a(s~;;;))
1/2
v(s)
+ 2v'a(s)p(s)h(t, s)
2
]
ds
To. Consequently,
li~~p H(:, T)
it
=v(T)-h~~f
1
H(t,T)
..
[H(t, s)Q(s) -
~a(s)p(s)h2(t, s)] ds
2
H(t,s) 1/2
1
]
[(a(s)p(s)) v(s)+2v'a(s)p(s)h(t,s) ds
59
for all T:::: To. Thus, by (2.5.60), we obtain
for all T:::: To. This shows that for T:::: To,
v(T) :::: Q(T)
(2.5.61)
and
t.-,oo
liminf
H t, T
r (()
'( ))
asp s
t
t~oo
<:::
H t, To
..
H (t s)
1/2
JT
Hence,
. . [ (1
hmmf
v(s)
+ -Va(s)p(s)h(t,s)
2
1t
v 2 (s()ds+
)
H(t,s) ()
(1 )
To
asp s
H t, To
hl::~f H(t, T~)
JTo
1t
To
ds <
00.
h(t,s)VH(t,s)v(s)dsJ
[(H(t'S))1/2
1
a(s)p(s)
v(s)+ 2va(s)p(s)h(t, s)J ds <
00,
(2.5.62)
i.e., we have
liminf [U(t)
t ..... oo
where
1
U(t)
H(t, To)
and
W(t) =
1t
1t
(1 )
H t, To
Now, we claim that
roo
JTo
~)
To
+ W(t)] <
H(t,s)
(2.5.63)
00,
v 2 (s)
a(s)p(s)
ds,
h(t, s)VH(t, s)v(s)ds,
t:::: To
t:::: To.
v 2 (s)
a(s)p(s) ds <
00.
v 2 (s)
a(s)p(s) ds =
00.
(2.5.64)
For this, suppose that
roo
JTo
(2.5.65)
By condition (2.5.57) there is a positive constant T/ such that

inf {lim inf H((t, s))}
t ..... oo H t, to
s"2to
> T/ > O.
(2.5.66)
Chapter 2
60
Let J.L be an arbitrary positive number. Then in view of (2.5.65) there

exists a T1 > To such that
h
t
To
Therefore, for t
U(t)
V2(S)
--:-:--,--:--:-ds
a(s)p(s)
T1 ,
1
H( t, 1'.)
0
>
To
H(t,s)d
for all
TJ
(h
T1 .
V2(T)
)
(
) ( )dT
a T p T
To
2(T) dT) ds
(r
iTo a(T)p(T)
2(T) dT) ds
rt -~H(t ' s) (r
iTo a(T)p(T)
H(t, To) iT! as
1
H(t, To)
>
ht
J.L
iTo
-~H(t
as
'
s)
( ~)
TJ
H(t, To)
iT!
-~H(t
as'
s)ds =
(~)
TJ
H(t, T 1)
H(t, To) .
By (2.5.66), there exists a T2 ~ Tl such that H(t, Td/ H(t, to) ~ TJ for
all t ~ T2, which implies that U(t) ~ J.L for all t ~ T2. Since J.L is
arbitrary
lim U(t) = 00.
(2.5.67)
t->oo
Next, consider a sequence {tk}k=l in (to, (0) with limk->oo tk = 00
satisfying limk->oo[U(tk) + W(tk)] = lim inft->oo[U(t) + W(t)]. In view of
(2.5.63), there exists a constant M such that
U(tk)
+ W(tk)
for
k = 1,2, .. '.
(2.5.68)
It follows from (2.5.67) that

00,
(2.5.69)
(2.5.70)
and hence (2.5.68) gives

lim W(tk) =
k->oo
00.
Taking into account (2.5.67), from (2.5.69), we get

1 + W(tk)
U(tk)
< ~ < ~
-
U(tk)
provided k is sufficiently large. Thus, W(tk)/U(tk) < -1/2 for all large
k, which by (2.5.70) ensures that
. w2(tk)
I1m
-::-::--:-'--:-'k->oo U(tk)
00.
(2.5.71)
61
On the other hand, by Schwartz's inequality, we have
But (2.5.66) guarantees that lim inft--->oo H(t, To)/ H(t, to) > TJ. This means
that there exists a T3 ~ To such that H(t, To)/ H(t, to) ~ TJ for every
t ~ T3. Thus, H(tk, To)/ H(tk, to) ~ TJ for all large k, and therefore,
W2(tk)
1
itk
U( ) ::;
H(
)
a(s)p(s)h2(tk' s)ds for all large n.
TJ
tk, to to
tk
=
(2.5.72)
00.
This gives limsuPt--->oo(l/H(t, to Itto a(s)p(s)h2(t, s)ds =

tradicts (2.5.58). Thus, (2.5.64) holds. Hence, by (2.5.61),
00
To
(D+(s2
a(s)p(s)
~--:-'-7-'-:-- ds
<
00
To
v 2(s)
ds <
a(s)p(s)
00,
which con-
00,
which contradicts condition (2.5.59). This completes the proof.
Consider the function H(t, s) = (t - s)n for t ~ s ~ to, where

n > 1 is a constant. Then from Theorem 2.5.12 the following corollary is
immediate.
Corollary 2.5.10. Let n> 1 be a constant and there exist two functions
p(t) E Cl([to,oo),lR+) and D(t) E C([to,oo),lR) such that condition
(2.5.59) hold,
limsuP~lt(t-s)n-2a(s)p(s)dS
n
t--+oo t
and for all T
to
<
00
to,
2
limsup : i t [(t - s)nQ(s) - n (t - s)n-2a(s)p(S)] ds > D(T).

t--+oo t to
4
Then equation (2.1.1) is oscillatory.
Chapter 2
62
Theorem 2.5.13. Let Hand h be as in Theorem 2.5.8 and conditions (2.5.46) and (2.5.47) hold. Suppose there exist two functions p(t) E
C 1([to, 00), JR+) and D(t) E C([to, 00), JR) such that condition (2.5.59)
hold,
liminf
T:::::
and for every
t-+oo
(1
H t, to
) i t H(t, s)Q(s)ds < 00
to,
lim inf H( 1 ) rt [H(t, s)Q(s) -
iT
t, T
t-->oo
(2.5.73)
to
~h2(t,
s)a(s)p(s)]
4
ds ::::: D(T)j (2.5.74)

Proof. Assume to the contrary that equation (2.1.1) is nonoscillatory. As
in Theorem 2.5.12, we obtain for t > T::::: To,
H(:, T) ht [H(t, s)Q(s) -
v(T) -
~a(s)p(s)h2 (t, S)] ds
H(:, T) l [(a~i;f:J 'I' *) + i via(s)p(, )h(t, s)
d,
Consequently,
lil~~f H(:,T) ht [H(t,S)Q(s) - ~a(s)p(s)h2(t,S)] ds
~ v(T) - H:':.':.::p H(:, T) l [(a~\t;(;)) 'I'v(,) + ~ viat' )p(, )h(t,
sf
for all T::::: To. It follows from (2.5.73) that
v(T) :0> f!(T)+
H:':.':.::p H (:, T l [ (.~\t;f:)) '/~W ~ via(,) p(,) h(t,
d,
,r
d'
for all T::::: To. Hence, (2.5.61) holds for all T::::: To and
.
iI~~p H(t, To)
t[(H(t'S))1
iTo
a(s)p(s)
/2
v(s)
+ 2yia(s)p(s)h(t, s)
]2 ds
::; v(To) - O(To) < 00.

This implies that
lim sup [U(t)
t-+oo
::;
iI~~p
+ W(t)]
H(t,s) 1/2
H(t, To)ho (a(s)p(s)) v(s)
1
+ 2yia(s)p(s)h(t, s)
ds < 00,
(2.5.75)
Oscillation and nonoscillation of linear ordinary diHerential equations
63
where U(t) and Wet) are defined in the proof of Theorem 2.5.12. By
condition (2.5.74),
n(to)
< liminf H( 1 )
t--->oo
liminf H( 1
:::::
[H(t,S)Q(S) -
t, to ltD
t--->oo
~h2(t,s)a(s)p(s)]
4
ds
t H(t,s)Q(s)ds
t, to ltD
-~4 lim
inf H( 1 ) rt h 2(t, s)a(s)p(s)ds.
t--->oo
t, to lto
From the above inequality and (2.5.73), we have
liminf H( 1 )
t--->oo
t h2(t,s)a(s)p(s)ds <
t, to ltD
00.
Thus there exists a sequence {tdk=l in (to,oo) with limk--->oo tk

satisfying
lim H( 1
k--->oo
tk, to lto
00
h 2(tk, s)a(s)p(s)ds
liminf H( 1
t--->oo
t, to lto
(2.5.76)
h2(t,s)a(s)p(s)ds <
00.
Now, suppose (2.5.65) holds. Using the procedure of Theorem 2.5.12, we

conclude that (2.6.69) is satisfied. It follows from (2.5.75) that there exists
a constant M such that (2.5.68) is fulfilled. Then as in Theorem 2.5.12, we
find that (2.5.72) holds, which contradicts (2.5.76). Hence, (2.5.65) fails.
Since the remainder of the proof is similar to that of Theorem 2.5.12, we
omit the details.
(t>'x'(t))'+w' cost)x(t)
for
t:2:to>O,
(2.5.77)
where A and I-l are constants, A::::: 1, - 1 < I-l ::::: 1 and 21-l + 1 :2: A.
Let pet) = 1 and H(t, s) = (t - s)2 for t:2: s :2: to. Then, we have
limsuPt--->oo(1/t 2 ) ft:(t - s)2sILcossds = -tbcosto < 00, and
liminf;
t--->oo t lT
[(t-S)2SIL coss-s>']ds :2: -TIL sinT-k 1 for T:2:to,
where kl is a positive small constant. Set neT) = -TIL sinT - k 1 .

Then there exists an integer N such that (2N + 1)1f + (1f/4) > to
and if n :2: N,
(2n + 1)1f + (1f/4) ::::: T ::::: 2(n + 1)1f - (1f/4), and
64
Chapter 2
= -TP.
D(T)
sin T - kl
2J-t + 1 ~ -X, we obtain
8TP., where 8 is a small constant. Noting that
L
00
82
00
s2P.- A ds
(2n+I)7r+(7r /4)
N=n
>
12(n+I)-(7r /4)
82
1 2 (n+I)-(7r/4)
ds
00.
(2n+I)7r+(7r/4) S
N=n
Thus, all conditions of Theorem 2.3.13 are satisfied, and hence equation
The above oscillation criteria as well as most of the known results in the
literature require information of equations (2.1.1) and (2.2.1) on the entire
half-line [to, 00). Now, motivated by the Sturm separation theorem, we will
present oscillation results for equations (2.1.1) and (2.2.1) which depend on
the behavior of the coefficients of these equations only on a sequence of
subintervals {[ai, bi]}~l of [to, 00), no matter how 'bad' the coefficients
are on the remaining parts of [to, 00 ).
In what follows, we say that the function H = H(t, s) E r if H is as
in Theorem 2.5.8, i.e., if HE C(V,lRf), where V = {(t, s): t ~ s ~ to}
which satisfies (2.5.46) and has partial derivatives oH/ot and oH/os on
V such that
otH(t,s) = hl(t,s)..jH(t,s)
osH(t,s)
and
- h2(t, s)..j H(t, s),

(2.5.78)
where hl,h2 E LLoc(V,JR).
From the results of Section 2.2, since equation (2.1.1) is equivalent to

(2.2.6), equation (2.2.6) is nonoscillatory if and only if (2.1.1) is nonoscillatory. As earlier, let wet) be a solution of (2.2.6) and assume that
wet) > 0 for t ~ T ~ to. For any pet) E CI([to, 00), JR+), we define
vet) = a(t)p(t)w'(t)/w(t), t ~ T. Then, it follows that
,
v (t)
v 2 (t)
+ Q(t) + a(t)p(t)
= 0
for
t ~ T,
(2.5.79)
where Q(t) is defined in (2.2.7).

The following lemma plays a fundamental role in establishing the interval oscillation criteria for equation (2.1.1).
Lemma 2.5.3. Let wet), vet) and T be defined as above.
(i)
Assume [c, b] C [T, 00). Then for any HEr,
Ib
c
H(b, s)Q(s)ds :S H(b, c)v(c)
+~
4
Ib a(s)p(s)h~(b,
c
s)ds.
(2.5.80)
Ic
65
Assume [a,c] C [T,oo). Then for any HEr,
(ii)
H(s, a)Q(s)ds :S - H(c, a)v(c)
1Ic
+4
a(s)p(s)hi (s, a)ds. (2.5.81)
Proof. (i) Multiplying equation (2.5.79) by H(b, s), integrating it with

respect to s from e to b and using (2.5.46) and (2.5.78), we get
jcbH(b, s)Q(s)ds
Jb
H(b, e)v(e) H(b, e)v(e)
JbcH(b, s)v'(s)ds - jbcH(b, s) a(s)p(s)

v2
(s)
ds
+~
jb H(b, s) a(s)p(s)
v2
(s)
ds
h2(b, s)JH(b, s)v(s)ds -
Jb h~(b, s)a(s)p(s)ds
c
))1/2 v(s) - 2Ja(s)p(s)h2(b,s) ]2 ds

j b[(H(b
a(s)~~s)
:S H(b, e)v(e) + ~ jb h~(b, s)a(s)p(s)ds.
4
-
(ii) Similar to part (i), we multiply (2.5.79) by H(s, a), integrate it with
respect to s from a to e and use (2.5.46) and (2.5.78), to get
Ic
H(s, a)Q(s)ds
- H(e, a)v(c)
=
Ic
c
+I
+4
1I
- H(e, a)v(e)
H(s, a)v'(s)ds -
v 2 (s)
H(s, a) a(s)p(s) ds
hies, a)a(s)p(s)ds
-t [C7~;~~)) 'I'
vi')
1Ic
a
v (s) ds
H(s, a) a(s)p(s)
hI (s, a)JH(s, a)v(s)ds a
< - H(e, a)v(e) + 4
Ic
Ic
+ ~v'a(S)p(')h'(S,a)r ds
hies, a)a(s)p(s)ds.
Corollary 2.5.11. Let wet), vet) and T be defined as above. Then for
any HEr and a, b, c E lR such that T:S a < e < b,
H(~, a)
<
H(s, a)Q(s)ds
~ [H(~,a) l
+ H(~, e)
lb
hi(s,a)a(s)p(s)ds +
H(b, s)Q(s)ds
H(~,c) lb h~(b'S)a(S)p(S)dS].
(2.5.82)
66
Chapter 2
Proof. Dividing (2.5.80) and (2.5.81) by H(b, c) and H(c, a) respec

tively, and adding them, we get (2.5.82).
Now, we shall prove the main interval oscillation theorems.

Theorem 2.5.14.
If for each T ~ to there exist HEr, p E
C 1 ([to, (0), JR+) and a, b, c E JR such that T <::: a < c < b, and
1 a)
H(c,
H(s, a)Q(s)ds
1
> 4"1 [ H(c,a)
+ H(b,1 c)
lb
C
2
hl(s,a)a(s)p(s)ds
H(b, s)Q(s)ds
1
+ H(b,c)
lb
c
2
]
h2(b,s)a(s)p(s)ds
,
(2.5.83)
then equation (2.l.1) is oscillatory.
Proof. Without any loss of generality we assume that w(t) is an eventually positive solution of equation (2.2.6). Then from Corollary 2.5.11 there
exists T ~ to such that inequality (2.5.82) holds for any HEr and
a, b, c E lR satisfying T <::: a < c < b. But, this contradicts (2.5.83).
Theorem 2.5.15. If there exists HEr such that for any r
t H(s, r)Q(s)ds
hm sup - " t - ' - T - - - - - t~=
and
IT hi(s, r)a(s)p(s)ds
1t H(t, s)Q(s)ds
Ian sup ---;-t-'-T _ _ _ _ __

t-->=
IT h~(t, s)a(s)p(s)ds
to,
00
(2.5.84)
00,
(2.5.85 )

Proof. For any T ~ to let a
there exists c > a such that
= T. In (2.5.84) we choose r = a. Then
11
H(s, a)Q(s)ds > 4"
In (2.5.85) we choose
lb
= c.
hi(s, a)a(s)p(s)ds.
(2.5.86)
Then there exists b> c such that
lib
H(b, s)Q(s)ds > -
h~(b, s)a(s)p(s)ds.
(2.5.87)
Combining (2.5.86) and (2.5.87), we obtain (2.5.83). The conclusion now

follows from Theorem 2.5.14.
The following result is an equivalent version of Theorem 2.5.15.
67
Theorem 2.5.16. If there exists HEr such that for any r 2: to,
limsuPjt [H(S, r)Q(s) t->oo
and
li~~p
it
[H(t,s)Q(S) -
~hi(s,
r)a(s)p(s)]
4
ds > 0
(2.5.88)
~h(t'S)a(s)p(s)]
ds > 0,
(2.5.89)

Proof. For any T 2: to let a = T. In (2.5.88) we choose r = a. Then
there exists c > a such that
[H(S, a)Q(s) -
~hi(s, a)a(s)p(s)] ds
> O.
(2.5.90)
In (2.5.89) we choose r = c. Then there exists b> c such that
Ib
[H(b,S)Q(S) -
~h~(b,S)a(S)p(s)] ds
> O.
(2.5.91 )
Combining (2.5.90) and (2.5.91), we obtain (2.5.83). The conclusion now

In the case H = H(t - s) E r, we have h1(t - s) = h 2 (t - s) and

denote them by h(t - s). The subclass of r containing such H(t - s)
is denoted by r o.
Theorem 2.5.14 when applied to ro leads to the following result.
Theorem 2.5.17. If for each T 2: to there exist two functions HEro
and p(t) EC1([to,oo),JR+) and a, cEJR such that T~a<c, and
1ic
H(s-a)[Q(s)+Q(2c-s)]ds > 4"
h2(s-a)[a(s)p(s)+a(2G~s)p(2c-s)]ds,
(2.5.92)

Proof. Let b = 2c - a. Then, H(b - c) = H(c - a) = H ((b - a)/2) ,
and for any y E .e[a, b], we have
y(s)ds =
y(2c - s)ds. Hence,
J: H(b -
lb
s)Q(s)ds =
J; H(s -
J:
J;
a)Q(2c - s)ds and
a(s)p(s)h2(b - s)ds =
a(2c - s)p(2c - s)h 2(s - a)ds.
Thus (2.5.92) implies that (2.5.83) holds for HEro, and therefore equation (2.1.1) is oscillatory.
68
Chapter 2
Now, let H(t - s) = (t - s)n where n> 1. It is clear that HE fo

and h(t - s) = n(t - s)(n-2)/2. Then from the above results the following
oscillation criterion is immediate.
Corollary 2.5.12. Equation (2.1.1) is oscillatory provided for some n> 1
and for any r::::: to either
(I)
the following inequalities hold

limsupit [(s - r)nQ(s) - n 2 (s - r)n-2 a(s)p(s)] ds > 0
t--->oo r
4
and
limsupit [(t - s)nQ(s) - n 2 (t - s)n-2 a(s)p(S)] ds > 0,
t--->oo r
4
or
(II)
the following inequality holds

lim sup
t--->oo
t [ (s
- r)n(Q(s)
x (a(s)p(s)
n
+ Q(2t - s - -(s
- r)n-2
4
+ a(2t - s)p(2t - s] ds >
O.
I:
Again, we define R(t) =

du/a(u) for t::::: r ::::: to, and let H(t, s) =
(R(t) - R(sn for t::::: s ::::: to, where n> 1 is a real constant. Then by
Theorem 2.5.16, we have the following result.
Corollary 2.5.13. Equation (2.1.1) is oscillatory provided limt--->oo R(t) =

for each r::::: to, and for some n > 1 the following inequalities hold
li~~p Rn-l(t)
it
n2
(R(s) - R(rnq(s)ds > 4(n - 1)
(2.5.93)
li~~p Rn~l(t)
it
(R(t) - R(snq(s)ds >
4(nn~ 1)'
(2.5.94)
00
and
Proof. Let p(t)
= 1.
Then, Q(t)
h 1 (t, s)
and
= q(t), t::::: to,
~(R(t) - R(s(n-2)/2
a(t)
~(R(t) - R(s(n-2)/2.
a(s)
69
Next, since
limsup
t~oo
n!l()]t [(R(S) - R(r)r'-lq(s) t
]t l(R(s)-R(r))nq(s)
r.
n
- T(R(s)-R(r))n-l a(s)
]t (R(s) - R(r))nq(s)ds
Rn-l( )
1
= li~~p Rn-l(t)
= lim sup
t~oo
- lim
t~oo
~a(s)hi(s,
r)] ds
4
2
1 ]
ds
Rn
!l( ) [4 (n2 1) (R(t) - R(r))n-l]

t
= lim sup Rn-l( )

t~oo
]t (R(s) - R(r)r'q(s)ds 17, -
17,2
( _ )
4 n
in view of (2.5.93) inequality (2.5.88) holds. Similarly, (2.5.94) implies that

(2.5.89) holds. Now, by Theorem 2.5.16 equation (2.l.1) is oscillatory.
Remark 2.5.8.
l. Special cases of the above oscillatory interval criteria for equation (2.l.1)
when a(t) = 1 and/or p(t) = 1 can be easily formulated. The details are
left to the reader.
2. Extensions of the above results to more general equations of type (2.2.1)
can be easily formulated (see Remark 2.5.6).
The following examples illustrate how the theory can be applied in practice.
x"(t)
+ q(t)x(t) =
0,
(2.5.95)
where q(t) is defined by
q(t) =
5(t-3m) for 3m:St:S3m+l

{ 5( -t + 3m + 2) for 3m + 1 < t :S 3m + 2
-m for 3m + 2 < t < 3m + 3,
where m E IN. For any T 2:: 0 there exists m E lN such that 3m 2:: T.
Let a = 3m, c = 3m+l, H(t-s) = (t-s?, t 2:: s 2:: T and p(t) = 1 for
t 2:: T. Then, Q(t) = q(t) for t 2:: T. It is easy to check that inequality
(2.5.92) holds, and hence every solution of equation (2.5.95) is oscillatory
by Theorem 2.5.14. We note that in this equation fooo q(s)ds = -00.
Example 2.5.8. Consider the Euler differential equation (2.2.11) with

t 2:: 1 and c> 1/4. Clearly, R(t) = fIt ds = t-l and limt~oo R(t) = 00.
70
Chapter 2
t:::: r
Since for each

lim
t~=
Rn
1 )
-l(
t
it
r
:::: 1 and n > 1,
(R(s) - R(r)tq(s)ds
it
it
1
lim R n -l( t )
t~=
c lim -1-
rs n -
t~oo
tn -
and
1 )
lim R -l(
n
t
t~oo
it
r
c
lim - (s - rt 2s ds :::: c t~=
tn - 1
2 -
nTs n - 3 ] ds
tn -
it
r
c hm
--1
tn -
t~oo
tit] ds
t n-l [ - - s2
s2
ds
n- 1
t~oo
sn - nrs n - 1
-C-
(R(t) - R(s))nq(s)ds
. -1c hm
it
. [t-
c Inn
t~oo
it
r
tn
nst n s
t]
- 1 - n In r
ds
oc.
condition (2.5.93) holds if c/ (n-1) > n 2 /[4( n-1 )]. Also condition (2.5.94)
is automatically fulfilled. Thus, Corollary 2.5.13 implies that the equation
(2.2.11) is oscillatory when c> 1/4.
2.6. Oscillation Criteria-Integrable Coefficients

In this section we shall present oscillation criteria for equation (2.1.1)
when the coefficient is integrably small. For this, we shall assume that
zo(t) E C([to, 00), JR) is a given function and there exists a function p(t) E
C 2 ([to,00),lR+) such that
00
/
--:-,.....--:---:-ds
a(s)p(s)
and
f3(t)
loo
where once more Q(t) = p(t)[q(t)
-p'(t)/(2p(t)).
(2.6.1)
00
Q(s)ds < 00,
+ a(t)h 2 (t)
(2.6.2)
- (a(t)h(t))'] and h(t) =
We define a sequence {zn(t)};;,,=o for t:::: to as follows (if it exists)
Zn(t)
zo(t)
00
(Z~_1 (s))2
a(s)p(s) ds,
= 1,2,
Oscillation and nonoscillation of linear ordinary diHerential equations
71
where z+(t) = (z(t) + Iz(t)I)/2. Clearly, Zl(t) 2 zo(t) and this implies
that zi(t) 2 z6(t). Thus, by induction, we have
Zn+l(t) 2 zn(t),
(2.6.3)
t2to
n=1,2,,
i.e., the sequence {zn(t)} is nondecreasing on [to, 00).

Theorem 2.6.1. Suppose zo(t)::; f3(t) for t 2 to. If equation (2.1.1) is
nonoscillatory, then there exists a tl 2 to such that
lim zn(t) = z(t) < 00
for
n->oo
2 tl.
(2.6.4)
Proof. Suppose equation (2.1.1) is nonoscillatory. Then in view of Theorem 2.2.3 there exists v(t) E C([tl, 00), IR) such that
v(t) = f3(t) +
00
v2(S)
a(s)p(s) ds
on [tl,oo) for some tl 2 to Thus, v(t) 2 zo(t) for t 2 tl, and hence
v+(t) 2 z6(t) for t 2 tl. This implies
f3(t)
v(t)
>
+.t
zo(t)+
00
V2(S)
a(s)p(s) ds 2 zo(t) +
1=
t
(Z6(8))2
() ()d8 = Zl(t)
asp.5
00
for
(v+(s)J2
a(s)p(s) ds
t2tl.
Hence, by induction, we have
v(t) 2 zn(t),
= 0,1,2,
for
(2.6.5)
t 2 tl.
Now, it follows from (2.6.3) and (2.6.5) that the sequence

bounded above on [tl,oo). Therefore, (2.6.4) holds.
{zn(t)}
is
From Theorem 2.6.1 the following oscillation result is immediate.

Theorem 2.6.2. Suppose zo(t)::; f3(t). If either
(i)
there exists a positive integer m such that zn(t) is defined for
n= 1,2,,m-1, but zm(t) does not exist, or
(ii) zn(t) is defined for n = 1,2, .. , but for arbitrarily large T* 2 to
there is t* 2 T* such that
00,
then equation (2.1.1) i::; oscillatory.

The following result guarantees the nonoscillation of (2.l.1).
(2.6.6)
72
Chapter 2
Theorem 2.6.3. Suppose zo(t):::: 1,B(t)1 for t:::: to. If there exists a
tl :::: to such that (2.6.4) holds, then equation (2.1.1) is nonoscillatory.
Proof. From (2.6.3) and (2.6.4) it follows that zn(t) ~ z(t), n = 0,1,
2, . .. for t:::: tl. Applying the monotone convergence theorem, we find
z(t) = zo(t)
Thus, we have
00
z+(t) = z(t)
>
00
00
(z+(s))2
a(s)p(s) ds :::: 0 for
(Z+(S))2
( ) (S ) ds
asp
00
+ Zo (t ) ::::
(z+(s))2 ds + ,B(t) I
a(s)p(s)
for
t:::: t I
(Z+(S))2
( ) (s ) ds
asp
+ I,B (t ) I
t:::: ti.
Now, it follows from Corollary 2.3.16 that (2.1.1) is nonoscillatory.
The next result provides necessary conditions for the oscillation of equation (2.1.1).
Theorem 2.6.4. Suppose zo(t):::: 1,B(t)1 for t:::: to. If equation (2.1.1)
is oscillatory, then either
(i)
there exists a positive integer m such that Zn (t) is defined for
n = 1,2,, m - 1, but zm(t) does not exist, or
(ii) zn(t) is defined for n = 1,2,, but for arbitrarily large T*:::: to
there is t*:::: T* such that (2.6.6) holds.
If ,B(t) :::: 0 for t:::: to,
corollaries are immediate.
then from the above results the following
Corollary 2.6.1. Suppose zo(t) = ,B(t) :::: 0 for t:::: to. Then equation
(2.1.1) is nonoscillatory if and only if there exists a h :::: to such that
(2.6.4) holds.
Corollary 2.6.2. Suppose zo(t) = ,B(t) :::: 0 for t:::: to. Then equation
(2.1.1) is oscillatory if and only if either
there exists a positive integer m such that zn(t) is defined for
(i)
n=1,2,,m-1, but zm(t) does not exist, or
(ii) zn(t) is defined for n = 1,2,, but for arbitrarily large T* :::: to
there is t*:::: T* such that (2.6.6) holds.
XIl(t)+~C4/3(2+cost+3tsint)x(t)
0,
t>O.
(2.6.7)

Let p(t)=1. Then, q(t)
Now, we let
zo(t)
= Q(t) = (1/3)C 4 / 3 (2+ cost +3tsint), t>O.
['>0 q(s)ds
f3(t)
73
= e l / 3 (2 + cost) > e l / 3 ,
t> O.
Then, zo(t)=zt(t) for all large t, andwehave
(z+(s))2
o
ds >
a(s)p(s)
-
00
and hence
2.6.2.
Tn
00
s-2/3ds =
00,
1. The oscillation of (2.6.7) now follows from Theorem
ExaIllple 2.6.2. Consider the differential equation
~ (e 3/4sin
x" (t) +
Here, we take p(t)
zo(t) = f3(t) =
~e5/4 cos
0)
> o.
(t-3/4 sin 0 + ~t-5/4 cos 0),
t>
x(t)
0,
(2.6.8)
1. Then,
q(t) = Q(t) =
Now, we let
1
~ 1=
00
o.
q(s)ds
[S-3/4 sin v's +
~s-5/4 cos v's] ds = e
l / 4 cos Vi.
Thus, zt(t) = (1/2)t-l/ 4 [cos 0 + I cos 01], and for k = [[tll + 1,
zI(t) =
11 Vs1
1
> -
1[(2n+3)7r/2]2 [ 1
00
2n=k
00
[(4n-l)7r/2J2
r57r /2
L j:{37r/2
00
[(2n+l)7r/2J2
r;:;
VS
(cos 2
j.[(4n+l)7r/2J2 cos2 Vs
n=k
[cos 2 v's + cos v'sl cos v'sl] ds + zo(t)
00
"2
Vs
ds
cos 2 sds + zo(t)
v's +
]
cos v'sl cos v'sl) ds + zo(t)
+ zo(t)
00.
n=k
Hence, all conditions of Theorem 2.6.2(i) are satisfied with

therefore (2.6.8) is oscillatory.
Tn
= 1,
and
74
Chapter 2
Now, we shall consider equation (2.1.2). Suppose Qo(t) E C([to, (0), IR)
is a given function and there exists a function p( t) E C 2([ to, 00 ), IR+) such
that
Ih(t)
1=
Q1(s)ds
and
00,
where Q1(t) is defined in (2.2.9).

Again, we define the sequence
exists)
Qn(t)
Qo(t)
00
(Q~_1(S))2
a1
for t 2 to as follows (if it
{Qn(t)}~=o
()
() ds,
s PI s
17, = 1,2,,
t :;;, to.
(2.6.9)
Clearly, Ql(t):;;, Qo(t) and this implies that Qi(t) :;;, Qt(t), t:;;, to Now,
by induction, we have
Qn+l(t) :;;, Qn(t),
t:;;, to
17, = 1,2,,
(2.6.10)
i.e., the sequence {Qn(t)} defined by (2.6.9) is nondecreasing on [to,oo).

From Theorems 2.6.1 and 2.6.3, we obtain the following comparison
result.
Theorem 2.6.5.
o<
Assume that
a(t)p(t) ::;; adt)Pl(t),
IfJl(t)l::;; f3(t)
for all large
t. (2.6.11)
If equation (2.1.1) is nonoscillatory, then (2.1.2) is nonoscillatory, or equivalently, if equation (2.1.2) is oscillatory, then (2.1.1) is oscillatory.
Proof. Let zo(t) = f3(t) and Qo(t) = 1f31(t)1 for t:;;, to. Suppose
equation (2.1.1) is nonoscillatory. It follows from Theorem 2.6.1 that there
exists a t1 2 to such that (2.6.4) holds. Clearly, by (2.6.11), Qo(t) =
1f31(t)1 ::;; f3(t) = zo(t), and hence Qt(t)::;; zt(t) for t 2 t 1. Thus, for
Qo(t)
< Qo(t)
1 t)
+ 1 a(~)p(s)
+
00
(Q+(S))2
()ds
a1
00
s PI s
(0+(S))2
ds
Now, by induction, we have
Qn(t) ::;; zn(t),
17,=0,1,2,,
t2 t 1
(2.6.12)
Therefore, by (2.6.4), (2.6.lO) and (2.6.12), we get
Q(t)
lim Qn(t) <
n----+CXl
lim zn(t)
n-+oo
= z(t) <
00,
t:;;, iI.

Hence, by Theorem 2.6.3, equation (2.1.2) is nonoscillatory.
75
Theorem 2.6.6. Suppose zo(t)::::: (3(t), t 2 to. If equation (2.1.1) is

nonoscillatory, then
(j
li~~p (z(t) - (3(t))exp
(3+(s)
)
a(s)p(s) ds
<
(2.6.13)
00,
where z(t) is defined in (2.6.4).

Proof.
Assume that equation (2.1.1) is nonoscillatory, then it follows
from Theorem 2.2.3 that v(t) = (3(t) + y(t) for t 2 T 2 to, where
y(t) = ftOO v 2(s)j(a(s)p(s))ds. Now,
yl (t)
a(t)p(t)
We claim that
I
Y (t)
:::
((3(t) + y(t))2
a(t)p(t)
v 2 (t)
(3+ (t)
+ 4 a(t)p(t) y(t) :::::
for
t 2 T.
(2.6.14)
t 2 T.
0,
Since y(t) >

and yl(t) < 0, (2.6.14) holds if (3(t) ::::: 0. If (3(t) 2 0,
then [y(t) + (3(tW 24{3+(t)y(t). This implies that (2.6.14) holds.
Clearly (2.6.14) implies that
y(t) ::::: y(T) exp ( -4
ht ats~~~~)
On the other hand, we have v(t) = (3(t)
y(t) =
00
v(t) = (3(t)
+ y(t) >
+ y(t) 2
v2(s)
a(s)p(s) ds >
-
This gives
zo(t)
00
t 2 T.
dS) ,
00
(2.6.15)
(3(t) 2 zo(t), and hence
t) ()
(Z+(S))2
asps ds.
(z+(s))2
a(~)p(s) ds
Zl(t)
for
t2T
and by induction
v(t) = (3(t)
+ y(t) 2 zn(t),
n = 0, 1,2,,,,,
t 2 T.
Therefore, from (2.6.15) and (2.6.16), we find
zn(t) - (3(t) ::::: y(t) ::::: y(T) exp
(3+(s)
)
(-4 Jrt a(s)p(s)
ds ,
T
(2.6.16)
Chapter 2
76
and hence
( t
,8+(s)
)
(Zn(t) - ,8(t)) exp 4 iT a(s)p(s) ds
< y(T),
n=0,1,2,,
t ~ T.
This and (2.6.4) lead to
(z(t) - ,8(t)) exp

=
lim (zn(t) -
n-->oo
(4 [
a~s~~~~) dS)
,8(t)) exp (4 t ~~(~)) dS) ::; y(T) <
iT asp s
Therefore, (2.6.13) holds.
00
for t
~ T.
The following result is now immediate.

Theorem 2.6.7. Suppose zo(t)::; ,8(t). If either
(i)
zn(t) exists for n = 1,2, ... , m and

li~~p (zm(t) - ,8(t)) exp
(j
4
,8+(s)
)
a(s)p(s) ds
00,
or
(ii)
(2.6.4) holds and

li~~p (z(t) - ,8(t))exp
(4 j
,8+(s)
)
a(s)p(s) ds
00,

Theorem 2.6.8. Suppose zo(t)::; ,8(t), t ~ to. If
j oo
exp -4
j8 a(u)p(u)
,8+(u)
)
du ds <
00,
(2.6.17)
and there exists a nonnegative integer m such that
joo zm(s)ds
(2.6.18)
00,

Proof. Suppose to the contrary that equation (2.1.1) is nonoscillatory.
As in Theorem 2.6.6, we obtain
(Zn(t) - ,8(t)) ::; y(T) exp
(-4 [
a~s~~~~) dS),
n = 0,1,2"",
~ T.
(2.6.19)

Integrating (2.6.19) from T to t and letting t
JTroo zn(s)ds
:::; y(T)
we find
--+ 00,
f3+(u)
)
JTr= exp (-4 JtT a(u)p(u)
du
77
ds+
JTroo f3(s)ds
<
00,
To prove our next result we shall need the following lemma.

Lemma 2.6.1. Every nonoscillatory equation (2.1.1) has a solution x(t)
such that
ds/(a(s)x2(s)) < 00 and a nontrivial solution y(t) such
that
ds/(a(s)y2(s)) = 00.
Joo
J=
Theorem 2.6.9. If
J=
(J8
1
a(s)p(s) exp -4
f3+ (T)
)
a(T)p(T) dT ds <
00,
(2.6.20)

Proof. If (2.1.1) is nonoscillatory, then from Theorem 2.2.3 there exists a
T ::>: to such that
v(t)
f3(t)
1=
t
v2(s)
a(s)p(s) ds
for
t::>: T,
where v(t) = a(t)p(t)w'(t)/w(t) and w(t) is a nonoscillatory solution of

(2.2.6) satisfying
ds/(a(s)p(s)w 2(s)) < 00. Now, as in Theorem 2.6.6,
we define y(t) = oo v 2 (s)/(a(s)p(s))ds, and obtain (2.6.15). It follows
y(s)/(a(s)p(s))ds < 00, which implies
from condition (2.6.20) that
j'XJ k(s)v2(s)/(a(s)p(s))ds < 00, where k'(t) = l/(a(t)p(t)). Next, since
Joo
Jt
( 1 w(t))2
n w(T)
J=
(tJ
: :; (tJT
T
v(s) d)2
a(s)p(s) s
ds
)
k(s)a(s)p(s)
there exist positive constants co, Cl such that

for all large t. Thus, we have
oo
ds
a(s)p(s)w2(s)::>:
c6
J=
(tJT
k (S)v 2(s) )
a(s)p(s) ds ,
Iw (t) I :::;
Co exp (c 1 Jln k (t) )
1
a(s)p(s)ex P (- 2c IJlnk(s))ds
00,
Corollary 2.6.3. If
1
(Js a(T)p(T)
f3(T)
)
Joo a(s)p(s)
exp -6
dT ds
<
00
for some 6, 0 < 6 :::; 4,

(2.6.21)
Chapter 2
78

ExaITIple 2.6.3. Consider (2.3.1) with c> 1/4. Let p(t) = tln 2A t,
where the constant >. is such that (1 - J4c - 1)/2 < >. < 1/2. Then, we
have
00
1
2c - 2>.2
2>' -1 < -1,
a(s)p(s) ds = 00,
1 2(A-1) t
Q(t) = (>.2 - A + c) n t '
fJ(t) =
00
and for T > 0,
1
T
A21-=-~: c ln2a - 1 t
Q(s)ds =
(l
>0
<
00
fJ(y) dy ) ds
a(y)p(y)
1 (2c-2A 2 )/(2A-1)
(ln 2(C+A 2 -A)/(2A-1)T)
n
s
s ds <
1
---:-:--exp
a(s)p(s)
00
- 2
roo
iT
00.
Now, it follows from Corollary 2.6.3 with 8 = 2 that (2.3.1) is oscillatory

if c> 1/4.
Corollary 2.6.4. Let k(t) E C 1 ([to,
If
00),
:JR+) satisfy k'(t)
l/(a(t)p(t)).
1
lim inf k(t)fJ(t) > -4'
(2.6.22)
t~oo

Proof. In view of (2.6.22) there exist two numbers ). > 1/4 and T 2: to
such that k(t)fJ(t) > A for t 2: T. Then, we have
00
1 - exp
-a(s)p(s)
<
(-4l
fJ+(y) dy ) ds
a(y)p(y)
(l
S
1
1
exp -4),
dy ) ds
a(s)p(s)
T a(y)p(y)k(y)
4A
1
(k(T))
a(s)p(s)(k(s))4A ds < 00.
00
roo
iT
ExaITIple 2.6.4. Consider the differential equation
((tln 2 t)x'(t))'
+ ~x(t) =
for
t 2: 2,
(2.6.23)
79
where c> 1/4 is a constant. Let p(t) = InA t and k(t) = _In- 1- At/(l +
A), where V4c -1 < A < -1. Then, we have
f3( ) = _ A2
t
+ 2A + 4c I l+A
+ A) n t
..
()
and hR:~f k t f3(t)
4(1
A2 + 2A + 4c
1
4(1 + A)2 > 2'
Now, it follows from Corollary 2.6.4 that (2.1.1) is oscillatory if c> 1/4.
Corollary 2.6.5. Let k(t) E C1([to, (0), JR+) satisfy k'(t)
If
liminf 4 ( ~ ( ) [4f3+(t)
t-HXl
a t p t
+ (a(t)p(s))']
>
= l/(a(t)p(t)).
~,
4
(2.6.24)

Proof. In view of (2.6.24) there exist two constants T;::: to and c> 1/4
such that
f3+(t)
c
(a(t)p(t))'
> - for t;::: T.
a(t)p( t)
t
4a( t )p( t)
Then, we have
00
1
--:-:--:--:exp
a(s)p(s)
<
(1
-4
f3+(T) dT ) ds
a(T)p(T)
roo a(s)p(s)
1
exp ( t [_ 4c + (a(T)p(T))'] d~) ds
J
T
a(T)p(T)
, .
JT
T4c
{(Xl
JT
a(T)p(T)
s- 4c ds < 00.
Example 2.6.5. Consider the Euler equation (2.2.11) with c> 1/4. Let
p(t) = tA, where A < 1 is a constant. Then, we have
f3(t) = A2 - 2A + 4c t A- 1
,
4(1 - A)
and
lil~~f 4a(t~p(t)
[4f3+(t)
t ;:::
+ (a(t)p(t))']
T> 0
4~~ =~)'
Now, it follows from Corollary 2.6.5 that (2.2.11) is oscillatory if (c(A/4))/(1- A) > 1/4.
In equation (2.2.6) if Q(t) > 0 for t;::: to, then we have the following
result.
Theorem 2.6.10.
Suppose Q(t) > 0 for t;::: to
C1([to,00),JR+) satisfy k'(t) = l/(a(t)p(t)). If
limsup k(t)f3(t) > 1,

t--+(Xl
and let
k(t) E
(2.6.25)
80
Chapter 2

Proof. Suppose (2.1.1) is nonoscillatory. It follows from Theorem 2.2.3
that there exists a T 2: to such that v(t) = ,B(t) + OO v 2(s)/(a(s)p(sds,
where
Jt
w'(t)
v(t) = a(t)p(t) w(t)
(2.6.26)
and w(t) is a solution of equation (2.2.6) with w(t) > 0 for t 2: T. If

Q(t) > 0 for t 2: to, then the function a(t)p(t)w'(t) is strictly decreasing
and v(t) ~ ,B(t) for t ~ T. Also, it is easy to check that w'(t) > 0 for
t 2: T. Now, since
w(T)
w(t)
k(t )w' (t)
+ ItT
a(s)p(s)w'(s)ds
a(s)p(s)
k(t)w'(t)
> w(T) + a(t)p(t)w'(t)[k(t) - k(T)] > a(t) (t) (k(t) - k(T)

-
k(t)w'(t)
k(t)
it follows that
.
k(t)
li~~p k(t),B(t) S; li~~p k(t)v(t) < h~~p k(t) _ k(T)
Theorem 2.6.11. If
I,
00
,B(s) ds <
a(s)p(s)
00
(2.6.27)
and t;,(t) defined in (2.2.23) satisfies
00
(IS
t;, (T)
)
a(T)p(T) dT ds <
a(s)p(s) exp -4
00,
(2.6.28)

Proof. Assume to the contrary that (2.1.1) is nonoscillatory. Then, by
Lemmas 2.2.2 and 2.6.1, equation (2.2.6) has a nonoscillatory solution w(t)
satisfying
ds/(a(s)p(s)w 2 (s < 00. As in the proof of Theorem 2.2.4,
the function z(t) =
v 2(s)/(a(s)p(sds satisfies (2.2.22), where v(t) is
defined in (2.6.26). Define y(t) = OO z2(s)/(a(s)p(sds. Then, we have
Joo
y' (t)
!too
z2 (t)
a(t)p(t)
1
Jt
2,B( t )y( t)
a(t)p(t)
- a(t)p(t) [Z2(t)
1
+ 2(t;,(t) + ,B(ty(t) + t;.2(t)]
< - a(t)p(t) [4t;,(t) + 2,B(t)]y(t).
(2.6.29)
ro
Using (2.6.27) and (2.6.28) in (2.6.29), we obtain

00. Observe that
r a(s)p(s)
y(s) d
s
iT
iTr (iTt
C1
iT
1
a(t)p(t) exp
C1
C2
iT
and C2, k(s)
(j'l
)2
z(s)
a (s) p( s) ds::;
y(s)j(a(s)p(s))ds <
(3 (u)
) ) z2 (s )
-2.1to a(u)p(u) du dt a(s)p(s) I1[S, to]ds
k(s)z2(s)
a(s)p(s) I1[S, to]ds 2:
for some positive constants

as in (2.2.24). Now, since
(j ry
(t
81
k(S)Z2(S)
a(s)p(s) ds
J; dtj(a(t)p(t))
k(S)z2(S) )
( ) ( ) ds
asps
(jTl
and I1[S, t]
1
)
( ) ( ) ( ) ds
ksasps
it follows that
iT
z(s)
a(s)p(s) ds ::; c3Vlnk (1])
for some
> O.
C3
(2.6.30)
By Schwartz's inequality and integration by parts, we have
(iTr
d)
v( s)
a(s)p(s) s
(iTr
::;
k(s)v 2 (s) ) (
a(s)p(s) ds
iT
1
)
k(s)a(s)p(s) ds
(2.6.31 )
and
fry k(s)v 2 (s)

a(s)p(s) ds = - k(1])z(1])
iT
fry
+ k(T)z(T) + iT
z(s)
a(s)p(s) ds.
(2.6.32)
From (2.6.30) - (2.6.32) the following estimate on the nonoscillatory solution w(t) is immediate
w(1]) _
In w(T) -
r a(s)p(s)
v(s)
ds
iT
<
-
c4ln
3/4
(2.6.33)
k(1]),
where C4 is an appropriate positive constant depending on

consequence of (2.6.33), we have
1
a( 1] )p( 1])w 2 (1])
C5
>
exp
(-2c
In 3 / 4 k(1]))
a(1])p(1])
exp (- In k (1] ) )
C6
a(1])p(1])
C6
k(1])a(1])p(1])
T.
As a
Chapter 2
82
for some positive constants
C5
1
d TJ
/ 00 a(TJ)p(TJ)w
2 (TJ)
---,--,---,.......,.---;:--:-c-
and
>
C6.
C6
Thus,
/00
1
d TJ
a(TJ)p(TJ)k(TJ)
00,
Finally, we present the following result.

Theorem 2.6.12. If equation (2.1.1) is nonoscillatory and
2
/ 00 a(s)p(s) exp (/s
(3(T)
)
a(T)p(T) dT ds =
00,
(2.6.34)
Joo a(s)u 2 (s)ds
then (2.1.1) cannot have a solution u(t) satisfying
<
00.

2.2.3 that there exist aT?: to and a function v(t) E C([T, 00), IR)
such that v(t) = (3(t) + oo v 2 (s)j(a(s)p(s))ds, where v(t) is defined in
(2.6.26) and w(t) is a solution of (2.2.6). Hence, for all t?: T, we have
Jt
w'(t)
w(t)
dt In w(t)
v(t)
a(t)p(t)
---'a(:-7t)l---'p(-:-t) [(3(t)
100 at:)~ls)
(3( t)
?: a(t)p(t) .
dS]
This implies
/=
a(s)u 2 (s)ds =
?:
/00 a(s)p(s)w
/=
(s)ds
a(s)p(s) exp
(2 /s a(~~:iT) dT) ds
00.
2.7. Forced Oscillations

In this section we are concerned with the oscillatory behavior of forced
equations of the form
(a(t)x'(t))'
+ p(t)x'(t) + q(t)x(t) =
e(t),
(2.7.1)
where a(t) E C([to,oo),IR+), and e(t), p(t), q(t) E C([to,oo),IR). In

what follows we shall assume that the unforced equation
(a(t)z'(t))' + p(t)z'(t) + q(t)z(t)
(2.7.2)
83
is nonoscillatory. Let z(t) be a nonprincipal solution of (2.7.2), i.e., z(t)

satisfies
00
1
- . , - -_ _-=--c--,- ds < 00,
/
(2.7.3)
a(s)p(s)z2 (s)
where
p(t)
(Jt ~~:~ dS)
exp
(2.7.4)
(see Lemma 2.6.1). Define the function
Jt a(s)p(~)Z2(s) (JS e(u)p(U)Z(U)dU) ds.
1jJ(t) =
(2.7.5)
Suppose equation (2.7.2) is nonoscillatory and let z(t)

be a nonprincipal solution. Then, (2.7.1) is oscillatory if
TheoreITl 2.7.1.
limsup 1jJ(t)
t-+oo
= -liminf 1jJ(t) =
t-+oo
Proof. The change of variable x(t)
(a(t)p(t)z2(t)y'(t))'
(2.7.6)
00.
= y(t)z(t) transforms (2.7.1) into
+ z(t) [(a(t)z'(t))' + p(t)z'(t) + q(t)z(t)] p(t)y(t)

= e(t)p(t)z(t),
where p(t) is defined in (2.7.4). We can express y(t)

(2.7.7) as follows
y(t)
= CrK::2
It
to
( ) ( ) 2( )ds+
asp s z s
It
( ) ( ) 2( )
asp s z S
to
(2.7.7)
by integration of
(is e(u)p(u)z(u)du)ds,
to
where C1 and C2 are constants depending on the initial conditions y(to)

and y'(to). Since z(t) is a nonprincipal solution, (2.7.3) and (2.7.6) imply
that y( t) satisfies
limsupy(t) =
t-+oo
-liminfy(t) =
t-+oo
00.
(2.7.8)
Because z(t) is nonoscillatory, z(t) of. 0 for t 2': t1 2': to. Now, (2.7.8)
implies that y(t) is oscillatory, and hence x(t) = y(t)z(t) is also oscillatory.
ExaITlple 2.7.1.
Consider the forced differential equation
x"(t)
where
0:,
+ sinkt x(t) =
t
(J, k are constants. For
z"(t)
Ikl > \1'2,
+ sin kt z(t) =
t
t 2': to > 0
t a sin(Jt,
(2.7.9)
the linear unforced equation

0
(2.7.lO)
84
Chapter 2
is nonoscillatory. Equation (2.7.10) has two linearly independent solutions

zd t) and Z'2 (t) with the following asymptotic behavior
Zl (t)
and
'""
(1'+(1/2)
Z2(t) '"" t~I'+(1/2)
,=
(1'+(1/2)[1
+ 0(1)]
C'"Y+(1/2) [1
+ 0(1)]
as
as
t ~ 00
t ~ 00,
where
((k 2 - 2)/(4k2))1/2 > O. Hence, Zj(t) is a nonprincipal
solution. Substituting zJ(t) and tOO sin;3t in (2.7.5), we find that (2.7.6)
is satisfied if Cl' >
(1/2) and any ;3 =/: O. Because, < 1/2, this shows
that equation (2.7.9) is oscillatory for any Cl' > 1 and for all ;3, k =/: O.
,+
Example 2.7.2. Consider the forced differential equation
xl/(t) - x(t) = e(t),
t 2 to 20
(2.7.11)
where e(t) E C([to, (0), JR). Here, a(t) = 1, p(t) = 0 and p(t) = l.
The linear unforced equation zl/(t) - z(t) = 0 is nonoscillatory. It has
two linearly independent solutions e~t and et . Clearly, z(t) = e t is a
nonprincipal solution.
Now, we consider the following two cases:
Let e(t) = sint Then, we have 1jJ(t) = te~2S (j'eTsinrdr)ds.
(i)
Clearly, 11jJ(t)1 < 00 for all t 2 0, and hence condition (2.7.6) is violated.
In this case the general solution of equation (2.7.11) is x( t) = C1e~t + c2e t (1/2) sin t, where Cj and C2 are arbitrary constants. It is nonoscillatory
for all sufficiently large t and C2 =/: o.
(ii)
Let e(t) = t 6e t sint or e<isint, where /5 > 0 and E> 1
are constants. Substituting z(t) = e t and e(t) in (2.7.5), we find that
(2.7.6) is satisfied. It follows from Theorem 2.7.1 that equation (2.7.11) is
oscillatory. Thus, the large size of the forcing term generates oscillation.
xl/(t)
+ 2x'(t) + x{t) =
sint,
t 2 to 2
o.
(2.7.12)
Here, a(t) = 1, p(t) = 2 and p(t) = e 2t . The linear unforced equation

zl/ (t) + 2z' (t) + z( t) = 0 is nonoscillatory. It has two linearly independent
solutions e~t and te~t. Clearly, z(t) = te~t is a nonprincipal solution. Now, since 1jJ(t) = t(1/s 2) (JsreTsinrdr)ds condition (2.7.6) is
satisfied. Hence, by Theorem 2.7.1, equation (2.7.12) is oscillatory.
The general solution of (2.7.12) is x(t) = C1e~t + c2te~t - (1/2)sint,
where C1 and C2 are arbitrary constants. It is oscillatory for all large t.
When condition (2.7.3) is violated, i.e., z(t) is not a nonprincipal
solution of equation (2.7.2), we have the following result.
85
TheoreIll 2.7.2. If there exists a positive solution z(t) of equation

(2.7.2) such that for each to> 0 and for some constant m> 0,
liminfit e(s)p(s)z(s)ds =
t----+oo
and
-00
to
t----+oo
!'IjI(t)! ~ m
and
lim
t->oo
lim sup
to
e(s)p(s)z(s)ds =
to
00,
(2.7.13)
to
it
( ) ( ) 2( )ds
asp
s z s
(2.7.14)
ds =
(2.7.15)
a(s)p(s)z2(s)
00,
where p(t) and 'IjI(t) are defined in (2.7.4) and (2.7.5) respectively, then
Proof. Let x(t) be an eventually positive solution of (2.7.1),say, x(t) > 0
for t 2: to > O. The function y(t) defined by x(t) = y(t)z(t) is a
nonoscillatory solution of equation (2.7.7). Integrating (2.7.7) from T> to
to t, where T is sufficiently large, we get
a(t)p(t)Z2 (t)y' (t)
= a(T)p(T)Z2 (T)y' (T) +
ht
e(s)p(s)z(s)ds
and hence by (2.7.13), we find lim inft->oo a(t)p(t)z2(t)Y'(t) =

we choose a sufficiently large tl 2: T such that
(2.7.16)
-00.
Next,
(2.7.17)
Replacing T by tl in (2.7.16), integrating it from tl to t and using
(2.7.14) and (2.7.17), we obtain
y(t) ~ y(tl) - m
t,
( ) ( ) 2( . ds
asp s z s)
and thus by condition (2.7.15), the solution y(t) is eventually negative.

This contradicts the fact that x(t) is eventually positive.
x"(t) - 3x'(t) + 2x(t)
e3t sint,
t 2: O.
(2.7.18)
The unforced equation Zll(t) - 3z'(t) + 2z(t) = 0 is nonoscillatory. It has

two linearly independent solutions et and e2t . Here, we let z(t) = et .
Since a(t) = 1 and p(t) = -3, we have p(t) = e- 3t . Thus, conditions
(2.7.14) and (2.7.15) are satisfied. It is easy to see that (2.7.13) also holds,
and hence, by Theorem 2.7.2 equation (2.7.18) is oscillatory. The same
86
Chapter 2
conclusion follows from Theorem 2.7.1 if we let z( t)

solution.
= e2t
as a non principal
The following corollary is now immediate.

Corollary 2.7.1. If there exists a positive solution y(t) of (2.7.2) such
that for each to > 0 conditions (2.7.6) and (2.7.13) are satisfied, then
Next, we consider a special case of (2.7.1), namely, the equation
(a(t)x'(t))'
+ q(t)x(t)
(2.7.19)
e(t)
and state the following result.

Theorem 2.7.3.
Let there exist two positive increasing divergent sequences {p~}, {p;;-}, and two sequences {c~} and {c;;-} such that
c~, c;;- are positive numbers, and
r::
Jp !;
+11" /,j";f
c;(l - q(s)) cos 2 R(t - p;)
+(q(s) - c;)
sin 2 (
V4(t -
p;))] ds
)
(2.7.20)
2 0
for all n 2 no, where no is a fixed positive integer. Further, let the
function e(t) satisfy
> 0 for
t E
[p~,p~ + 7r/.Jc;fl
:::: 0
t E
[p;;- ,p;;- + 7r/~1
e(t) { -
for
(2.7.21)
for all n 2 no. Then equation (2.7.19) is oscillatory.

The following result extends Theorem 2.7.3.
Theorem 2.7.4. Suppose for any T 20 there exist T::::
t2 such that
e(t)
{ <0
20
for
for
tE[sl,hl
t E
[S2, t2]'
Sl
< t1
:::: S2
<
(2.7.22)
If there exist
such that
(2.7.23)
for i
= 1,2
87
Now, we shall consider a more general forced equation
+ p(t)x'(t) + q(t)f(x(t))
(a(t)x'(t))'
where the functions a, e, p and

C(lR, lR), xf(x) > 0 for x -=I- O.
e(t),
(2.7.24)
are as in (2.7.1) and
f(x) E
The following oscillation criterion for equation (2.7.24) extends and

unifies Theorems 2.7.3 and 2.7.4. For this, as in Theorem 2.7.4 for each
T ::::: 0, T ::; SI < tl ::; S2 < t2, and i = 1,2 we shall need the set
D(Si,ti)'
Theorem 2.7.5. Let
f'(x) ::::: k > 0 for
x -=I- 0,
where k is a real constant,
(2.7.25)
and suppose for any T::::: 0, there exist T::; SI < tl ::; S2 < t2 such
that condition (2.7.22) holds. If there exist p(t) E Cl([Si,ti],lR+) and
u(t) E D(Si,ti) such that
Qi(U)
l~i p(s) [q(S)U
2 (S)
a~)
(u'(s)
+ 2a~\;(s)) 2]
ds ::::: 0
(2.7.26)
for i = 1,2 where
l'(t) = a(t)p'(t) - p(t)p(t),
(2.7.27)

Proof. Suppose x(t) is a nonoscillatory solution of equation (2.7.24),
say, x(t) > 0 for t::::: to ::::: O. Define w(t) = -p(t)a(t)x'(t)/ f(x(t)) for
t ::::: to. Then for t::::: to, we have
,
W
p(t)e(t)
(t) = p(t)q(t) - f(x(t))
l'(t)
w 2 (t)
+ a(t)p(t) w(t) + f (x(t)) a(t)p(t)'
(2.7.28)
By hypotheses, we can choose Sl, tl ::::: to, Sl < tl so that e( t) ::; 0

on I = [SI, tl]' Thus, on the interval I, w(t) satisfies the differential
inequality
,
w (t) ::::: p(t)q(t)
l'(t),
w 2 (t)
+ a(t)p(t) w(t) + f (x(t)) a(t)p(t)
on
I.
(2.7.29)
Let u(t) E D(sl, tt) be as in the hypothesis. Multiplying (2.7.29) by

u 2 (t) and integrating over I, we find
u 2 (s)w'(s)ds
:::::
u 2 (s)p(s)q(s)ds
1a(~~;~s)
r '(x(s))u (s) a(s)p(s)

w (s)
ds.
+ JI f
u 2 (s)w(s)ds
(2.7.30)
Chapter 2
88
Integrating the left-hand side of (2.7.30) by parts and using the fact that
u(sI) = U(t1) = 0, we obtain
-21 u(s)u'(s)w(s)ds ;:::
u 2(s)p(s)q(s)ds
J'
1a(;~;~s)
u 2(s)w(s)ds
2
w 2 (s)
If (x(s))u (s) a(s)p(s) ds,
which is equivalent to
;: : J[ a(s)p(s)
f'(x(s)) u(s)w(s) + (u'(s) + ')'(S)U(S))
2a(s)p(s)
Q1(U)
a(s)p(s)]
f'(x(s)) dS+Q1(U).
Since
2:::
f'(x(t)) u(t) (_ (t) a(t)x'(t))

f(x(t))
a(t)p(t)
p
i.e.,
[_f'(x(t))X'(t)u(t)
f(x(t))
or
(2.7.31)
inequality (2.7.31) yields
d ,
dt Y (t)
u'(t)
,),(t)u(t)
')'(t)
0,
2a(t)p(t)
,),(t)u(t)
+ 2a(t)p(t) =
+ 2a(t)p(t)y(t)
0,
+ 2a(t)p(t)
+ u'(t)] + ')'(t)u(t)
d ( u(t))
f(x(t)) dt f(x(t))
or
on
0,
I,
(2.7.32)
where yet) = u(t)/ f(x(t)), tEl. It follows from (2.7.32) that (e(t)y(t))'
on I, where
e(t) = exp
(1~ 2a!X~(s) dS),
tEl.
Thus, e(t)y(t) = C for some constant C, and hence u(t) = Cf(x(t))/e(t)

on I. Since u(t) E D(Sl' t 1 ) and u(t) =I- 0, this is incompatible to the
fact that x(t) > on I. This contradiction proves that x(t) is oscillatory.
When x(t) <

eventually, we use u(t) E D(s2' t2) and e(t) 2:::
[S2, t2J to reach a similar contradiction. This completes the proof.
on
Theorem 2.7.5 when specialized to (2.7.19) takes the following form:
Corollary 2.7.2. Suppose for any T;:::

there exist T:::: Sl < tl :::: S2 <
t2 such that condition (2.7.22) holds. If there exist pet) E C1([Si' til, JR+)
89
and u(t) E D(Si, t i ) such that
Qi(U)
lSift;
( ) u(s) )
p(s) [q(s)u 2(s) - a(s) (
u'(s) +'
;p(:)
2] ds::::
(2.7.33)
for i = 1,2 then equation (2.7.19) is oscillatory.
ReIllark 2.7.1. Theorem 2.7.4 follows from Corollary 2.7.2 by letting
p(t) = 1. Also, if the weight function
(2.7.34)
then "((t) defined in (2.7.27) will be identically zero. Thus, the condition
(2.7.26) in Theorem 2.7.5 takes the form
Qi(U) =
1:;
p(s) [q(s)u 2(s) -
a~) (U'(S))2] ds
:::: 0
(2.7.35)
for i = 1,2.
ExaIllple 2.7.5. Consider the forced equation
(ytx'(t))' +
~x'(t) + ~x(t)
sinVt,
> O.
(2.7.3b)
Here, the zeros of the forcing term sin Vt are (mr)2.

For any T> 0 choose n sufficiently large so that (mr)2:::: T, and
set S1 = (mr)2 and t1 = (n + 1)211"2 and take u(t) = sin Vt, p(t) = t
and k = 1. Then, we have
l~n+l)7r 2v2 (sin2 v - ~ cos 2 v) dv
(n+1)7r V2
-(3 - 5 cos 2v)dv

4
3n 2 + 3n + 111"3 _ ~11" > O.
4
16
n7r
Similarly, for S2 = (n + 1)211"2 and t2 = (n + 2)211"2 we can show that

Q2(U) > O. Now, it follows from Theorem 2.7.5 that equation (2.7.36) is
oscillatory.
Chapter 2
90

1. The results of Section 2.1 are due to Kreith [17,18J and Swanson [30J.
2. Lemma 2.2.1 is theorem 2.15 of Swanson [30], Lemma 2.2.2 can be

proved by Kummer transformation (see Willett [31]), i.e., for (2.1.1) we
let w(t) = x(t)/ viP(t), where p(t) E C2([to, 00), JR+). Theorem 2.2.6
improves a result of Hartman [12], Theorem 2.2.7 improves the comparison
theorem due to Hille [13J and Wintner [34], while Theorem 2.2.8 improves
a result of Wong [35J. In Theorem 2.2.1, we can let the function p(t) E
= exp (t(p(s)/a(s))ds). In this case Q*(t)

and h*(t) defined in (2.2.3) and (2.2.4), respectively, become h*(t) = 0
and Q*(t) = p(t)q(t).
C2([to,00),JR+) to be p(t)
3. Theorem 2.3.1, Corollaries 2.3.1, 2.3.4 and 2.3.7 are due to Wintner
[33], Kneser [15], Hille [13J and Moore [22], respectively. Corollary 2.3.6
with a(t) = 1 and c = 1/4 reduces to Wintner's criterion [34J. Also
Corollary 2.3.11 with a(t) = p(t) = 1 reduces to a result of Potter [27J.
Theorem 2.3.3 improves a result of Wong [35], whereas Corollary 2.3.17
improves a result of Wintner [33J. Corollary 2.3.19 is due to Moore [22J.
Corollary 2.3.20 with A = 2, p(t) = 1 and f3(t) ~ 0, reduces to Willett's
criterion [31J.
4. Theorems 2.4.1 and 2.4.3 are taken from Graef et. al. [9J. Theorem
2.4.1 extends Leighton's oscillation criterion established in [20J.
5. Theorem 2.5.1 is extracted from Grace and Lalli [8J, and it includes
Wintner result [32J. Theorem 2.5.2 is due to Graef et. al. [9J. Theorem
2.5.3 includes Hartman's oscillation criterion [11 J as a special case. Lemmas 2.5.1 and 2.5.2 and Theorem 2.5.4 extend Willett's oscillation result
[31]. Theorems 2.5.6 - 2.5.8 are taken from Kamenov [14], Yan [37J and
Philos [25], respectively. Corollaries 2.5.5 - 2.5.7 are due to Philos [23,24J.
Theorems 2.5.12 and 2.5.13 are the extensions of Philos results [25J and are
borrowed from Garce [7J. Lemma 2.5.3 and Theorems 2.5.14 - 2.5.17 are
the generalizations of the work originated by Kong [16].
6. Theorems 2.6.1- 2.6.4 extend the results ofYan [38J. Theorem 2.6.5
is due to Li [21]. Lemma 2.6.1 is taken from Coppel [5J. Corollary 2.6.3
includes Wintner's criterion [33J. Theorem 2.6.11 generalizes theorem 2 of
Wong [35J, and Theorem 2.6.12 improves Wintner result [32J.
7. Theorems 2.7.1 and 2.7.2 are the extensions of Rankin results [28J.
Theorem 2.7.3 is due to EI-Sayed [6], and Theorem 2.7.4 is taken from Wong
[36]. Theorem 2.7.5 improves Theorem 2.7.4, and is taken from Agarwal
and Grace [2J.
8. For some other related results of this chapter see Agarwal et. al.
[1,3J.
91
2.9. References
l. R.P. Agarwal and S.R. Grace, On the oscillation of certain second
order differential equations, Georgian Math. J. 7(2000), 201-213.

2. R.P. Agarwal and S.R. Grace, Second order nonlinear forced oscillations, Dyn. Sys. Appl., to appear.
3. R.P. Agarwal, S.R. Grace and D. O'Regan, Oscillation Theory for
Second Order Dynamic Equations, to appear.
4. R.P. Agarwal and R.C. Gupta, Essentials of Ordinary Differential
Equations, McGraw-Hill, Singapore, 1991.
5. W.A. Coppel, Disconjugacy, Lecture Notes in Math., 220, SpringerVerlag, New York, 1971.
6. M.A. El-Sayed, An oscillation criterion for a forced second order linear
differential equation, Proc. Amer. Math. Soc. 118(1993), 813-817.
7. S.R. Grace, Oscillation theorems for nonlinear differential equations of
second order, J. Math. Anal. Appl. 171(1992), 220-241.
8. S.R. Grace and B.S. Lalli, Integral averaging and the oscillation of
second order nonlinear differential equations, Ann. Mat. Pura Appl.
151(1988), 149-159.
9. J.R. Graef, S.M. Rankin and P.W. Spikes, Oscillation theorems for
perturbed nonlinear differential equations, J. Math. Anal. Appl. 65(1978),
375-390.
10. G.H. Hardy, J .E. Littlewood and G. Polya, Inequalities, 2nd ed.
Cambridge Univ. Press, 1988.
11. P. Hartman, On nonoscillatory linear differential equations of second order, Amer. J. Math. 74(1952), 389-100.
12. P. Hartman, Ordinary Differential Equations, John Wiley, New York,
1964.
13. E. Hille, Nonoseillation theorems, Trans. Amer. Math. Soc. 64(1948),
234-252.
14. LV. Kamenev, An integral criterion for oscillation of linear differential
equations of second order, lVlclt. Zametki 23(1978), 249-251.
15. A. Kneser, Untersuchungen uber die reellen Nullstellen der Integrale linearer Differentialglcichungen, Math. Annalen 42( 1893), 409-435.
16. Q. Kong, Interval criteria for oscillation of second order linear ordinary
differential equations, J. Math. Anal. Appl. 239(1999), 285-270.
17. K. Kreith, Oscillation Theory, Lecture Notes in Math., 324, Springer
Verlag, New York, 1973.
18. K. Kreith, PDE generalization of Sturm comparison theorem, lVIemoirs
Amer. Math. Soc. 48(1984),31-46.
19. M.K. Kwong and J.S.W. Wong, An application of integral inequality
to second order nonlinear oscillation, J. Differential Equations 46(1982),
63-77.
20. W. Leighton, The detection of the oscillation of solutions of a second
order linear differential equation, Duke J. Math. 17(1950), 57-62.
92
Chapter 2
21. H.J. Li, Oscillation criteria for second order linear differential equations,
J. Math. Anal. Appl. 194(1995), 217-234.
22. R.A. Moore, The behavior of solutions of a linear differential equation of
second order, Pacinc J. Math. 5(1955), 125-145.
23. Ch.G. Philos, Oscillation of second order linear ordinary differential equations with alternating coefficients, Bull. Austral. Math. Soc. 27(1983),
307-313.
24. Ch.G. Philos, On a Kamenev's integral criterion for oscillation of linear
differential equations of second order, Utilitas Math. 24(1983),277-289.
25. Ch.G. Philos, Oscillation theorems for linear differential equations of second order, Arch. Math. 53(1989), 482-492.
26. M. Picone, Sui valorieccezionali di un parametro de cui dipends un equazioni differenziale lineare ordinaria del second ordine, Ann. Scuola Norm.
Pisa 11(1909), 1-141.
27. R.L. Potter, On self-adjoint differential equations of second order, Pacinc
J. Math. 3(1953),467-191.
28. S.M. Rankin, Oscillation results for a nonhomogeneous equation, Pacinc
J. Math. 80(1979), 237-243.
29. B. Sturm, Sur les equations differentielles lineaires due second ordre, J.
Math. Pures Appl. 1(1836), 106-186.
30. C.A. Swanson, Comparison and Oscillation Theory of Linear Differential
Equations, Academic Press, New York, 1968.
31. D.Willet, On the oscillatory behavior of the solutions of second order
linear differential equations, Ann. Polon. Math. 21(1969), 175 194.
32. A. Wintner, A criterion for oscillatory stability, Quart. Appl. Math.
7(1949), 115-117.
33. A. Wintner, On the nonexistence of conjugate points, Amer. J. Math.
73(1951), .368-380.
34. A. Wintner, On the comparison theorem of Kneser-Hille, Math. Scand.
5(1957), 255-260.
35. J.S.W. Wong, Oscillation and nonoscillation of solutions of second order linear differential equations with integrable coefficients, Trans. Amer.
Math. Soc. 144(1969), 197-215.
36. J .S.W. Wong, Oscillation criteria for a forced second order linear differential equation, J. Math. Anal. Appl. 231(1999), 235-240.
37. J. Yan, Oscillation theorems for second order linear differential equations
with damping, Proc. Amer. Math. Soc. 98(1986), 276-282.
38. J. Yan, Oscillation property for second order differential equations with
an 'integrably small' coefficient, Acta Math. Sinica 30(1987), 206-215.
Chapter 3
Oscillation and Nonoscillation
of Half-Linear Differential
Equations
3.0. Introduction
In this chapter we shall present oscillation and nonoscillation criteria
for second order half-linear differential equations. In recent years these
equations have attracted considerable attention. This is largely due to the
fact that half-linear differential equations occur in a variety of real world
problems; moreover, these are the natural generalizations of second order
linear differential equations. In Section 3.1, we shall provide some preliminaries for the study of half-linear differential equations. In Sections 3.2 and
3.3, respectively, Sturm's and Levin's type comparison theorems are developed. In Section 3.4, we shall establish a Liapunov type inequality. Section
3.5 presents an oscillation criterion for almost periodic Sturm-Liouville
equations. A systematic study on the zeros of solutions of singular ha1linear equations is made in Section 3.6. Nonoscillation characterizations
(necessary and sufficient conditions), comparison results as well as several
sufficient criteria for the nonoscillation are presented in Section 3.7. Section 3.8 is devoted to the study of oscillation of half-linear equations. In
Section 3.9, we shall establish oscillation criteria by employing integral and
weighted averaging techniques. Here, interval criteria for the oscillation
of half-linear equations are also provided. Section 3.10 deals with the oscillation of half-linear equations with integrable coefficients. Section 3.11
addresses the oscillation of damped and forced equations. In Section 3.12,
we shall derive lower bounds for the distance between consecutive zeros of
an oscillatory solution. Finally, in Section 3.13, we shall present a systematic study of the oscillation and nonoscillation of half-linear equations with
a deviating argument. Here, classifications of the nonoscillatory solutions,
and the existence results which guarantee that the solutions have prescribed
asymptotic behavior are also presented.
94
Chapter 3
3.1. Preliminaries
In this chapter we shall study half-linear differential equations of the
form
d
(3.1.1 )
dt (a(t)'l/J(x'(t) + q(t)'Ij;(x(t = 0,
where a(t) E C([to,oo),JR+), q(t) E C([to,oo),JR) and 'Ij;: JR ---+ JR is

defined by ~jJ(y) = iyi",-ly with a > 0 is a fixed number. If a = 1,
then (3.1.1) reduces to the linear differential equation
d
dt (a(t)x'(t
+ q(t)x(t)
(3.1.2)
O.
A solution of (3.1.1) is a real valued function x(t) E C1([to, 00), JR). If

x(t) is a positive solution of (3.1.1) and either increasing, or decreasing,
then equation (3.1.1) reduces to the Euler-Lagrange equation
:t (a(t)[x'(t)]"') + q(t)x"'(t) = 0,
or
(3.1.3)
dt (a(t)[-x'(t)]"') - q(t)x"'(t)
(3.1.4)
respectively.
The existence of solutions to initial value problems for (3.1.1) on [T,oo),
T ~ 0 has been addressed in [15]. We note that any constant multiple
of a solution of (3.1.1) is also a solution. A solution x(t) of (3.1.1) is
called oscillatory if for each tl > 0 there exists a t > tl such that
x(t) = 0; otherwise, it is said to be nonoscillatory. Equation (3.1.1) is
called oscillatory if all its solutions are oscillatory, and nonoscillatory if
all its solutions are nonoscillatory. The main objective of this chapter is
to discuss the oscillation and nonoscillation behavior of all solutions of
equation (3.1.1).
3.2. Sturm-Type Comparison Theorems

In what follows AC(I) denotes the set of all absolutely continuous
functions on an interval I ~ JR. To prove our main results, we shall need
the following five lemmas.
Lemma 3.2.1 [23]. If A and B are nonnegative constants, then

AI..
and
AA
+ (A -
+ (A -
l)BA - AAB A- 1 ~ 0
l)BA - AAB A- 1
:s
for
for
A> 1
< A < 1,
Oscillation and non oscillation of half-linear differential equations

where equality holds if and only if A
95
= B.
Lemma 3.2.2. Let a(t) E C1((a,b),JR+) and q(t) E C((a,b),JR).

Suppose equation (3.1.3) has a positive increasing solution x(t) on (a, b).
If y(t) E AC((a,b),JRo) satisfies
liminf y<>+l(t) [x'(t)]<> a(t)

t-+bx( t)
then
l!~~~f
ld
limsup y<>+l(t) [x'((t))]<> a(t),

t-+a+
X t
[a(s)ly'(s)I<>+l - q(s)y<>+l(s)] ds
(3.2.1)
(3.2.2)
0,
d --+ b-
where equality holds if and only if x and yare proportional.

Proof. Define g(t) = x'(t)/x(t) on (a, b). It follows from (3.1.3) that
(a(t)gU(t))'
+ aa(t)gu+l(t)
q(t)
on
(a,b).
(3.2.3)
Let
a1/(u+ll(t)y(t)g(t)
and
), = a
+1
in Lemma 3.2.1, to obtain
where equality holds if and only if g(t) = y'(t)/y(t). Thus, we have

(3.2.4)
where equality holds if and only if x and yare proportional. Now,
integrating (3.2.4) from c(> a) to d( < b), we obtain
Id
[a(s)ly'(s)I<>+l-q(s)y<>+l(s)] ds
~ a(d)y<>+l(d)g<>(d)--a(c)y<>+l(c)gC>(c).
It follows from (3.2.1) that (3.2.2) holds.
Using an argument similar to that in Lemma 3.2.2, we can prove the

following lemma.
Lemma 3.2.3. Let a(t) E C1((a,b),JR+) and q(t) E C((a,b),JR).
Suppose equation (3.1.4) has a positive decreasing solution x(t) on (a, b).
If y(t) E AC((a, b), JRo) satisfies
liminf y<>+l(t) [x'(t)]

t-+a+
x(t)
<>
a(t)
limsup yo<+l(t) [x'(t)]

t-+bx(t)
0<
a(t),
Chapter 3
96
then (3.2.2) holds with equality only if x and y are proportional.

Lemma 3.2.4. Let a(t) E C1a,b),lR+) and q(t) E Ca,b),lR).
Suppose x(t) is a solution of equation (3.1.1) on [a, b) satisfying x(a) =
x(b) = 0, then x'(a) #- 0, or x'(b) #- O.
Proof. We shall prove only the case x'(a) #- O. Assume to the contrary
that x'(a) = O. Since any constant multiple of a solution of (3.1.1) is
also a solution of (3.1.1), we can assume without loss of generality that
x(t) > 0 on (a,b). It follows from equation (3.1.1) that a(t)'l/J(x'(t)) =
q(s)'l/J(x(s))ds on [a, bJ, which implies
-J:
x'(t)
'l/J-l ( - att)
it
q(S)'l/J(X(S))dS) ,
where 'l/J-l is the inverse function of 1jJ.

C1([a,bJ,lR+), we have
x(t)
< (t - a)'l/J-l
for
a::::
iT
(Mit
'l/J-l ( - atr)
Mit
q(S)'l/J(X(S))dS) dr
Iq(s)I'l/J(x(S))dS)
t :::: b, where M = max{l/a(t) : a
'l/J(x(t)) :::: (b - a)"
Since x(a) = 0 and a E
::::
t :::: b}. Hence,
Iq(s)I'l/J(x(s))ds
for
a:::: t :::: b.
Thus, it follows from Lemma 1.1.1 that 'l/J(x(t)) = 0 for each t E [a, b).
This implies that x(t) = 0 on [a, bJ, which contradicts the hypothesis
x(t) > 0 on (a,b). This contradiction completes the proof.
From Lemmas 3.2.2 - 3.2.4, we shall establish the following Wirtingertype inequality.
Lemma 3.2.5. Let x(t) be a solution of equation (3.1.1) on [a,b)
satisfying x(t) #- 0 on (a, b). Denote by Y the family
Y = {y(t) E C1([a,b),lR): y(a) = y(b) = 0 and y(t)
#- 0
on (a,b)}.
Then for every y E Y,

(3.2.5)
where equality holds if and only if x and yare proportional.
Osci11ation and nonoscillation of half-linear differential equations
97
Proof. Let y E Y We shall prove only the case y(t) > 0 on (a, b),
since the proof of the case y(t) < 0 on (a, b) is similar. Since x(t) is a
solution of equation (3.1.1) on [a, b) satisfying x(t) -=I- 0 on (a,b), we
may assume without loss of generality that x(t) > 0 on (a, b). Now, we
separate our proof into the following six cases:
Case 1. Either (i). there is a number c, a < c 0 on (a, b).
(i). It follows from y(a)
= 0, x'(c) = 0 and Lemma 3.2.4 that
lim y"+l(t) [x'((t))] " a(t) =
t~a+
=0
lim y"+l(t) [x'((t))] a a(t) = O. (3.2.6)
t~c-
The proof of (3.2.6) is as follows:

(I) If x(a) = 0, then x'(a) -=I- 0 by Lemma 3.2.4. This and y(a) = 0
imply that
lim y,,+l(t) [x'(t)]" a(t)
( lim y((t)))" lim y(t)a(t)
x(t)
Ha+
t~a+ X
t~a+
. y(t))"
( t~~+
x(t)
(II) If x'(a)
imply
= 0, then x(a)
0=
[Y'(a)]"
x'(a)
x 0 = O.
-=I- 0 by Lemma 3.2.4. This and
lim y,,+l(t) [x'(t)]" a(t)
x(t)
t~a+
(III) It follows from x'(c)
=0
y(a) = 0
= O.
and x(t) -=I- 0 on (a,b) that
lim y,,+l(t) [x'(t)]" a(t)
x(t)
He-
O.
From (I) - (III), (3.2.6) is clear.

Now, by (3.2.6) and Lemma 3.2.2, we have
a(s)ly'(s)I"+lds 2:
(ii). It follows from y(a)
= y(b) = 0
lim y"+l(t) [x'((t))] " a(t) =
t~a+
q(s)y"+l(s)ds.
lim y"+l(t) [x'((t))] " a(t)

x t
a(s)ly'(s)I"+lds 2:
(3.2.7)
and Lemma 3.2.4 that
t~b-
This and Lemma 3.2.2 now imply that
lb
lb
q(s)y"+l(s).
= O.
(3.2.8)
(3.2.9)
98
Chapter 3
Case 2. Either (iii). there is a number c, a < C +l (t)
Ha+
[_ x' (t)]
x(t)
<>
= 0, x'(c) = 0
a(t)
lim
=0
y<>+l (t)
He-
[_ x' (t)] <> a(t)

x(t)
o.
(3.2.10)
This and Lemma 3.2.3 imply that (3.2.7) holds.
(iv). It follows from y(a) = y(b) = 0 and Lemma 3.2.4 that
lim y<>+l(t) [- XI(t)]<> a(t) =
x(t)
Ha+
lim y<>+1(t) [- X'(t)]<> a(t) =

x(t)
Hb-
o.

Case 3. If there are two numbers Co and Cl, a < Co <
x'(co) = X'(Cl) = 0 and x'(t) > 0 on (co, Cl), then
lim y<>+l(t) [Xl(t)] 0: a(t) =

x(t)
HCci
lim yO:+1(t) [Xl(t)]

x(t)
Cl
<>
+1(t) [- x'((t))] <> a(t) =

x t
t ..... cci
Cl
+l(t) [_ Xl((t))] <> a(t)

x t
O.
t ..... c:;-
It follows from Lemma 3.2.2 that (3.2.11) holds.
Case 5. Either (v). there is a number c, a < C 0 on (c,b], or (vi). x'(t) > 0 on [a,b).
(v). It follows from y(b)
lim yO:+l(t) [XI(t)]
x(t)
<>
= 0, x'(c) = 0
a(t)
Hc+
lim yO:+1(t) [x'(t)]

x(t)
Hb-
=0
0:
a(t)
O.
This and Lemma 3.2.2 imply that

(3.2.12)
Oscillation and nonoscillation of half-linear differential equations
99
(vi). It follows from y(a) = y(b) = 0 and Lemma 3.2.4 that (3.2.8)
holds. This and Lemma 3.2.2 imply that (3.2.9) holds.
Case 6. Either (vii). there is a number c, a < c < b satisfying x'(c) = 0
and x'(t) < 0 on (c,b], or (viii). x'(t) < 0 on (a,b].
(vii). It follows from y(b)
= 0, x'(c) = 0 and Lemma 3.2.4 that
lim y"'+l(t) [- X'(t)] '" a(t):::::

x(t)
t--+c+
lim y"'+l(t) [_ x'(t)] '" a(t) = O.

x(t)
t--+b-

(viii). It follows from y(a) = y(b) = 0 and Lemma 3.2.4 that (3.2.10)
holds. This and Lemma 3.2.3 imply that (3.2.8) holds.
Combining the above six cases, we obtain the inequality (3.2.5).
Remark 3.2.1. If a is a quotient of two positive odd integers, then the

condition 'y(t) #- 0 on (a, b)' in Y can be omitted.
Now, we shall use Lemma 3.2.5, to prove the following Sturm-type

comparison theorem for equation (3.1.1).
Theorem 3.2.1. Let a(t), al(t) E C1([a,b],JR+) and q(t), ql(t) E
C([a, b], JR). If the boundary value problem
(al(t)1f;(y'(t)))'
+ ql(t)1f;(y(t)) =
y(a) = y(b) = 0,
has a solution y(t) with y(t)
lb
(3.2.13)
#- 0 on (a, b) satisfying
[(a(s) - al(s))ly'(s)I",+l - (q(s) - ql(s))ly(s)I"'+l] ds :S 0, (3.2.14)
then every solution x(t) of (3.1.1) must have a zero in (a, b) unless x(t)
and y(t) are proportional.
Proof.
Suppose to the contrary that x(t) is a solution of equation
(3.1.1) such that x(t) #- 0 on (a, b). It follows from Lemma 3.2.5 that
(3.2.5) holds. Multiplying equation (3.2.13) by y(t) and integrating it by
parts from a to b, we get
al(s)IY'(s)I<>+lds =
ql(s)ly(s)I<>+lds.
Combining this with (3.2.5), we find
J:
lb
J:
[(a(s) - al(s))ly'(s)I"'+l - (q(s) - ql(s))ly(s)I<>+l] ds ::::: 0,
which contradicts (3.2.14) unless equality holds. In view of Lemma 3.2.5

the later situation occurs if and only if x(t) and y(t) are proportional.
This completes the proof.
Chapter 3
100
Remark 3.2.2. If 0' = 1, then Theorem 3.2.1 reduces to the well-known

Sturm comparison theorem.
If a(t) = al(t)
following corollary.
and
q(t) = ql(t),
then Theorem 3.2.1 yields the
Corollary 3.2.1. The zeros of two linearly independent solutions of equation (3.1.1) sepiirate each other.
Remark 3.2.3. Corollary 3.2.1 implies that the oscillatory and nonoscillatoryequations (3.1.1) are mutually exclusive.
Corollary 3.2.2. Let a(t), al(t) E C1([to,oo),IR+) and q(t), ql(t) E
C([to,oo),IR). Suppose al(t) 2: a(t) and ql(t) S; q(t) on [to, (0). If the
differential equation
(3.2.15)
is oscillatory, then so is (3.1.1); or equivalently, the nonoscillation of (3.1.1)
implies that of equation (3.2.15).
3.3. Levin-Type Comparison Theorems

In this section we shall extend Levin's comparison theorem which is
known for the differential equation
xl/(t)
+ ql(t)X(t) =
and
yl/(t)
+ q2(t)y(t) =
to the half-linear differential inequality
x(t) ((al(t)'!j;(X'(t)))'
+ ql(t)'!j;(X(t))} <
(3.3.1)
and the half-linear differential equation

(3.3.2)
where al(t), a2(t) E C1([T,00),IR+), T 2: to 2: 0, ql(t), q2(t) E C([T,oo),
IR) and ql, q2 E .cloc([T,oo),IR), and al(t) S; a2(t) for all t 2: T.
Theorem 3.3.1. Let x(t) and y(t) be nontrivial solutions of (3.3.1)
and (3.3.2) respectively, on a closed subinterval [a, b] of [T, (0) satisfying
either
(i)
(ii)
y(a) 2: x(a) > 0 and x(t) > 0 on [a, b], or

y(a) S; x(a) < 0 and x(t) < 0 on [a, b].
If
'!j;(x'(a))
-al(a) '!j;(x(a))
it
a
'!j;(y'(a))
ql(s)ds > I-a2(a) '!j;(y(a))
it
a
q2(s)ds
(3.3.3)
101
on [a, b], then y(t) does not vanish on [a, b], and
y(t) > x(t) > 0

y(t) ::::: x(t) < 0
x(t)x'(t) < 0,
if (i) holds,
if (ii) holds
also
Ia2 (t) 1jJ (y' (t)) I
1jJ(x' (t))
-al(t) 1jJ(X(t)) >
If >
by
t E [a,b].
for all
1jJ(y(t))
(3.3.4)
in (3.3.3) is replaced by 2, then > in (3.3.4) should be replaced
2.
Proof. Since the proofs of (i) and (ii) are similar, we shall prove only case
(i). Since x(t) > 0 on [a,b], the continuous function
w(t)
satisfies
w'(t) 2 ql(t)
Thus, we find
w(t)
w(a)
w(a)
it
it
1jJ( x' (t))

al(t) 1jJ(X(t))
[a,b]
on
+ cw~l/"'(t)lw(t)l(o+l)/(l
ql (s)ds
it
(3.3.5)
2 ql(t).
oa-1/"'(s)lw(s)I(Q+l)/ads
(3.3.6)
ql(s)ds.
Hence, x'(t) < 0 for all t
z(t)
[a, b]. Since y(a) > 0 the function
= -
1jJ(y'(t))
a2(t) 1jJ(y(t))
is continuous on some interval [a, c],

satisfies the integral equation
where a
(3.3.7)
< c < b. Clearly, z(t)

(3.3.8)
for all t E [a, c]. Moreover, we have
z(t) 2 z(a)
it
q2(s)ds.
(3.3.9)
It follows from (3.3.3), (3.3.6) and (3.3.9) that
z(t) 2 z(a)
it
q2(s)ds > - w(a)
-it
ql (s)ds
> - w(t) on [a, c].

(3.3.10)
Chapter 3
102
Hence, w(t) > -z(t) on [a, e]. Next, we claim that w(t) > z(t) on [a, e].
For this, assume that there exists a tl E (a,e) such that w(tJ} = Z(tl)
and z(t) < w(t) on [a, tl]' By (3.3.10), we have 0 < jz(t)j < w(t) on
[a, tl]' Thus, it follows from (3.3.3), (3.3.6) and (3.3.8) that
Z(tl)
+ itt qz(s)ds + itt aa;-I/O:(s)jz(s)j(O:+l)/O:ds
z(a)
< w(a)
+ itt ql(s)ds + itt aa~I/O:jw(s)j(O:+I)/O:ds <
w(t l ),
which is a contradiction. Therefore,
jz(t)j < w(t)
on
[a, e].
(3.3.11)
Next, we shall show that y(t) cannot vanish on [a, b]. Suppose the
first point to the right of a at which y(t) vanishes is t = t2 :s; b, i.e.,
y(t) > 0 on [a, t2) and y(tz) = O. Then from Lemma 3.2.4 it follows
that y'(t2) =I O. This means that the solutions of equation (3.3.2) have only
simple zeros. However, since jz(t)1 < w(t) on [a, tz) and w(t) is bounded
on [a,b], we get 00 = limsuPHc jz(t)j :s; lim Ht - w(t) = W(t2) < 00,
which is a contradiction. Hence, 2y(t) cannot va~ish on [a, b]. Thus,
(3.3.11) holds on [a, e] C [a, b] on which z(t) is continuous. But, this
implies that z(t) is continuous on the entire interval [a, b], since w(t) is
bounded on [a, b] and y(t) cannot vanish on [a, bj. Therefore, (3.3.11)
holds on the interval [a, bj.
Now, from (3.3.5), (3.3.7) and (3.3.11) it follows that y(t) 2': x(t) on
[a, bj. This completes the proof.
Theorem 3.3.2. Let x(t) and y(t) be nontrivial solutions of (3.3.1)

and (3.3.2) respectively, on [a, bj c [T,oo) satisfying either
x(t) > 0 on [a,bj and y(b) 2': x(b) > 0, or

x(t) < 0 on [a, b] and y(b):S; x(b) < O.
(i)
(ii)
If
'ljJ(x'(b))
al(b) 'ljJ(x(b))
r ql(s)ds
+ it
'ljJ(y'(b))
> a2(b) 'ljJ(y(b)) + it qz(s)ds
(3.3.12)
for all t E [a, b], then y(t) does not vanish on [a, b], and
x(t)x'(t) < 0,
y(t) > x(t) > 0 if (i) holds,

y(t) < x(t) < 0 if (ii) holds
also
'ljJ(x'(t))
al(t) 'ljJ(x(t)) >
'ljJ(y'(t)) I
az(t) 'ljJ(y(t))
for all
[a, b].
(3.3.13)
Oscillation and nonoscillation of half-linear differentia.l equations

If > in (3.3.12) is replaced by
replaced by ~.
>
then
~,
103
in (3.3.13) should be
Proof. Define
Xl(t)
= x(a+b-t),
Al(t)
Yl(t)
al(a+b-t),
A2(t)
Ql(t) = ql(a+b-t),
Q2(t)
y(a+b-t)
= a2(a + b - t)
= q2(a + b - t).
Then, Xl(t) does not vanish on [a,b], Al(a) = al(b), Yl(a) = y(b)
x(b) = xl(a) > 0, or Yl(a) = y(b) ::::: x(b) = xl(a) < 0, and
'lj;(x~ (a))
-A1(a) 1p(xl(a))
a +b- t
Ib q2(s)ds I
1p(y(b)) +
Ia2(b) 'lj;(y'(b))
t
Ql(s)ds
=
'lj;(x' (b))
al(b) 1p(x(b))
I-A2(a) 'lj;(Yl(a))
'lj;(y~(a)) +
Ib
ql(s)ds,
I ba
Q2(s)ds .
Thus, (3.3.12) is equivalent to the assumption (3.3.3), since t E [a, bJ if

and only if a + b - t E [a, bJ. Hence the conclusion (3.3.13) follows from
Theorem 3.3.1.
Corollary 3.3.1. If the hypotheses of Theorems 3.3.1 and 3.3.2 satisfied,

then the following inequalities hold for t E [a, bJ,
'lj;(y'(t))
'lj;(x'(t))
a2(t) 'lj;(y(t)) - al(t) 'lj;(x(t))
'lj;(y'(a))
'lj;(x'(a))
> a2(a) 'lj;(y(a)) - al(a) 'lj;(x(a))
It
+
a
(3.3.14)
(ql(S) - q2(s))ds
and
'lj;(x'(t))
'lj;(y'(t))
al(t) 'lj;(x(t)) - a2(t) 'lj;(y(t))
'lj;(x'(b))
'lj;(y'(b))
> al(b) 'lj;(x(b)) - a2(b) 1p(y(b)) -
Ib
t
(3.3.15)
(ql(S) - q2(s))ds.
Proof. It follows from the proof of Theorem 3.3.1 that
z(t) - z(a)
<
-it
q2(s)ds =
it
aa;-l!a(s)lz(s)l(a+l)!ads
it aa~l!Q(s)lw(s)I(Q+l)!Qds
w(t) - w(a)
-it
ql(s)ds.
This in view of
z(t) =
'lj;(y'(t))
- a2(t) 'lj;(y(t))
and
w(t)
'lj;(x'(t))
- al(t) 'lj;(x(t))
104
Chapter 3
is equivalent to (3.3.14). Similarly, (3.3.15) can be obtained from (3.3.14)

by the substitution used in the proof of Theorem 3.3.2.
As applications of Theorem 3.3.1, we shall prove the following results.

Theorem 3.3.3. Suppose for some a
T,
(i)
x(t) and y(t) are nontrivial solutions of (3.3.1) and (3.3.2), respectively, and either
y'(a) = 0, x'(a) ::; 0, y(a) ~ x(a) >

(II) y'(a) = 0, x'(a) ~ 0, y(a) ::; x(a) <
(I)
(ii)
it
ql(s)ds
lit
for all t
Q2(s)dsl
Then, y(t) does not vanish for t
x(t)x'(t) < 0,
also
and
and
~
x(t) >
x(t) <
on
on
[a, (0), or
[a, (0),
a.
a, and
<
y(t) > x(t) >
if (I) holds,
y(t) < x(t)
if (II) holds
_al(t)1P(X'(t)) >
1p(X(t)) -
la (t)1j;(y'(t))
I
1j;(y(t))
for
~ a.
Proof. Since (i) implies - adt)1p(x'(a))j1j;(x(a)) ~ 0, (3.3.3) holds on

every closed interval [a,b]. Hence, the conclusion of Theorem 3.3.1 holds
on every such interval. This in turn implies that the conclusion of Theorem
3.3.3 holds on every such interval. This completes the proof.
Theorem 3.3.4. Suppose there exists a nontrivial solution y(t) of (3.3.2)

which satisfies the conditions y(a) = y(b) = 0, y'(e) = 0, a < e < b. If
the inequalities
hold on [a, e] and [e, b]

respectively, then every solution x(t)
(al(t)1j;(X'(t)))'
+ Ql(t)1j;(X(t)) =
of the
(3.3.17)
has at least one zero in [a, b].

Proof. Let x(t) be a nontrivial solution of (3.3.17) satisfying x'(e) = 0.
We assert that x(t) has at least one zero in each of the intervals [a, e)
and (e,b]. From Lemma 3.2.4 it follows that x(e) f.
and y(e) f. 0.
Without loss of generality we can assume that x(e) = y(e). If x(t) has
no zero in (e, b], and hence no zero on [e, b], (3.3.16) ensures that all the
Oscillation and nonoscilla.tion of half-linear differential equations
105
hypotheses of Theorem 3.3.1 are satisfied. Therefore, y(t) has no zero in

[c, b]. This contradicts the hypothesis y(b) = O. Similarly, if x(t) has no
zero in [a, c) an application of Theorem 3.3.2 yields the contradiction that
y(t) has no zero in [a, c]. Hence, x(t) has at least two zeros in [a, b],
and therefore by Corollary 3.2.1 every solution of equation (3.3.17) has at
least one zero in [a, b].
3.4. Liapunov's Inequality

The well-known Liapunov inequality can be stated as follows:
Theorem 3.4.1. Let q(t) E C((a,b),lR+) be not identically zero on any
open subset of (a, b). Suppose x(t) is a nontrivial solution of equation
+ q(t)x(t)
x"(t)
having consecutive zeros at t
holds
= a and t = b. Then the following inequality
(b-a)
lb
q(s)ds > 4.
In this section, first we shall extend Liapunov's inequality to the halflinear differential equation
+ q(t)~/J(x(t)) =
('l/J(x'(t)))'
(3.4.1 )
and then present an application.

Theorem 3.4.2. Let q(t) E C([a,b],lRo) be not identically zero on any
open subset of [a, b]. Suppose x(t) is a nontrivial solution of equation
(3.4.1) having consecutive zeros at t = a and t = b. Then the following
inequality holds
Proof. Since for the equation (3.4.1) any constant multiple of a solution
x(t) is also a solution, we may assume without loss of generality that
x(t) > 0 on (a, b). Let M be the maximum of x(t) on [a, b] which is
attained at some point to = a + f,(b - a), 0 < f, < 1. Then, we have
'l/J(x'(a)) - 'l/J(x'(b))
d
- Jar dt'l/J(x'(t))dt
1
b
q(t)'l/J(x(t))dt < MC<
1
b
(3.4.2)
q(t)dt.
It follows from x(t) > 0 and q(t) :2': 0 that x(t) is concave. Hence, we find
x'(a) :2': M/[f,(b - a)] and x'(b):::; -M/[(1 - f,)(b - a)]. This and (3.4.2)
Chapter 3
106
f:
imply that C'" + (1 - ~)-'" < (b - a)'"

q(t)dt. On the interval (0,1)
the function ~-'" + (1 - ~)-'" has a maximum 2",+1 which is attainded
at ~ = 1/2. Therefore, 2",+1 < (b - a)'"
q(t)dt. This completes the
proof.
f:
Theorem 3.4.3. Let q(t) E C([a, b], JR). Suppose x(t) is a solution of
equation (3.4.1) having consecutive zeros at t = a and t = b. Then the
following inequality holds
2"'+1
< (b - a)'"
lb
q+(t)dt,
(3.4.3)
where q+(t) = max{q(t),O}.

Proof. Consider the differential equation
(1jJ(y'(t)'
+ q+(t)1jJ(y(t
(3.4.4)
O.
The existence theory ensures that the initial value problem (3.4.4), y(a) =
0, y'(a) = 1 has a solution y(t). Then from Theorem 3.2.1 it follows that
y(t) has at least one zero in (a, b]. Let ~ = min{ t E (a, b] : y(t) = O}. Thus,
from Theorem 3.4.2, we have (b - a)'"
q+(t)dt 2: (~- a)'" q+(t)dt >
2"'+1. This completes the proof.
f:
f:
Suppose x(t) is a nontrivial solution of (3.4.1) having N + 1 consecutive

zeros at a = ao < a1 < ... < aN = b. Then from (3.4.3), we have
k=1,2,,N.
(3.4.5)
Oscillation and nonoscillation of half~linear differential equations
107
Therefore, we have 2::=1 (ak - ak~l)~<> 2:: N<>+l (b - a)~<>. This and
(3.4.5) imply that
q+(t)dt > (2N)<>+!(b - a)~<>. From this inequality
the following corollary is immediate.
J:
Corollary 3.4.1. Let q(t) E C([a, b], lR) be not identically zero on [a, b].
Suppose x(t) is a nontrivial solution of equation (3.4.1) having consecutive
zeros at a = ao < a1 < ... < aN = b. Then the following inequality holds
lb
q+(t)dt > (2N)<>+!(b -
a)~<>.
Definition 3.4.1.
A function f E C(lR, lR) is said to be quickly
oscillatory if it is defined in a neighborhood of 00 and there exists a
sequence {tn}~=l with limn-+oo tn = 00 such that f(t n ) = 0, n =
1,2,, tn+ 1 > tn and limn-+oo(t n+ 1 - tn) = O.
As an application of Theorem 3.4.3, we shall prove the following result.
If equation (3.4.1) has a quickly oscillating solution,
Theorem 3.4.4.
then
(3.4.6)
and limsuPt-+oo q(t) =
00.
Proof. Let x(t) be a quickly oscillating solution of equation (3.4.1) with

zeros tn such that tn --+ 00 and tn+! - tn --+ 0 as n --+ 00. Consider
the consecutive zeros tn+! > tn > to and the interval [tn' tn+d. Then it
follows from Theorem 3.4.3 that
00
q+(t)dt >
to
It
q+(t)dt >
2<>+!(tn+1-tn)~<>
--+
00
as n--+oo.
tn
Hence (3.4.6) holds. Now, from the mean value theorem for integrals, we
q+(t)dt = q+(~n)(tn+1 - tn) > 2<>+1(tn+! - tn)~<>' where
have
tn < ~n < tn+!. This implies that q+(~n) > 2<>+!(tn+! - tn)~<>~l and
hence limsuPt-+oo q(t) = 00. This completes the proof.
JLn+l
3.5. An Oscillation Criterion for AlmostPeriodic Sturm-Liouville Equations

The class n c .cLc(lR) of Besicovitch almost periodic functions is the
closure of the set of all finite trigonometric polynomials with the Besicovitch
seminorm I . I B defined by
IlqllB =
1
lim sup -2
t-+oo
jt
~t
Iq(s)lds,
108
Chapter 3
where q E fl. The mean value, M {q} of q E fl always exists, is finite

and is uniform with respect to f3 for f3 E JR, where
it
M{q} = lim -1
t .... oo
to
q(s + (3)ds = 0
to 2:
for some
o.
Consider the following half-linear differential equation
('ljJ(x'(t)))' - )..q(t)'ljJ(x(t))
where q(t) E fl, ).. E JR - {O} and
~/J(x)
0,
(3.5.1)
is as in equation (3.1.1).
Definition 3.5.1. We say equation (3.5.1) is oscillatory at + 00 and

- 00 if every solution of (3.5.1) has an infinity of zeros clustering only at
+ 00 and - 00, respectively.
When
0: =
1 equation (3.5.1) reduces to the linear differential equation
x"(t) - )..q(t)x(t)
From Levin's comparison theorem with
immediate.
O.
0: =
(3.5.2)
1 the following result is
Theorem 3.5.1. Let q(t) Efland M{lql} > O. Then, M{q} = 0 if

and only if equation (3.5.2) is oscillatory at + 00 and - 00 for every
).. E JR - {a}.
In what follows we shall extend the sufficient condition of Theorem 3.5.1
to equation (3.5.1). For this, we need the following two lemmas.
Lemma 3.5.1. If I> 1, a, bE JR+ then (a
+ b)'Y >
a'Y
+ b'Y.
Proof. Without loss of generality we can assume that a 2: b. Then by

Lemma 1.1.1, we have
Lemma 3.5.2. Suppose

{
q: [to, (0) -+ JR is locally Lebesgue integrable,

to 2: 0 and has the mean value M {q} = O.
(3.5.3)
If x(t) -=f. 0 is a solution of equation
('ljJ(x'(t)))' - q(t)'ljJ(x(t))
on [to, (0), then
. lit (IX'(S)I)"'+l
Ix(s)1
hm -
Hoo
to
--
ds
o.
(3.5.4)
109
Proof. Define w(t) = -'lj;(x'(t))N(x(t)) for t;::: to. It follows from

(3.5.4) that w(t) is a solution of the equation
w'(t) - o:lw(t)l(o+l)/o
+ q(t)
lim sUPt-+oo (1!t)
It suffices to show that
to the contrary that
lit Iw(s)l(o+l)/Ods
t
t;::: to.
It: Iw(s)l(o+l)/ods =
limsup t-+oo
for
>
to
(3.5.5)
O. Assume
o.
(3.5.6)
Integrating (3.5.5) from to to t and dividing it by t, we obtain
w(t)
w(to)
t
+~
t
it q(s)ds + ~t it Iw(s)l(o+l)/Ods
to
for t> to.
to
(3.5.7)
From (3.5.6), (3.5.7) and (3.5.3) there exist a positive constant m and an
increasing sequence {t n };:"=l of [to, (0) with limn -+ oo tn = 00 such that
w(t n ) > o:m o+1

tn
for all sufficiently large n.
(3.5.8)
It follows from (3.5.3) that there exists t* large enough so that
Ii:
it
for all
q(s)dsl <
t ;::: t*.
(3.5.9)
Using (3.5.9) for t;::: tn ;::: t*, we find
it
q(s)ds =
tn
q(s)ds
to
_ltn
to
q(s)ds < 200:+ 1 m 0+1( t
+ tn ) .
(3.5.10)
From (3.5.8) and (3.5.10), we get
w(t n )
-it
tn
q(s)ds > o:mo+ltn
2~1 mO+l(t + tn)
> o:mo +1t n - ~m<>+l

[(2<>+1 - l)t n + t]
2<>+1
n
for all tEl = [tn, (2 0 +1
(3.5.11)
l)t nl C [t*, (0).
From the existence theory the differential equation
!('lj;(x~(t))) -
0:
(;)"+1 'lj;(Xn(t))
has a solution xn(t) on I satisfying xn(tn) = x(t n ), and

'lj;(X~ (in))
'lj;(Xn(t n ))
(3.5.12)
l10
Chapter 3
Now, from (3.5.10) and (3.5.11), we obtain
Thus, from Theorem 3.3.1 it follows that

(3.5.13)
Now, define wn(t) = -1jJ(x~(tn))N(xn(tn)) on I. It follows that wn(t)
is a solution of the differential equation
(3.5.14)
(3.5.15)
and
(3.5.16)
on h = [tn, tn +cnl c [t*, (0), where n is large enough so that wn(tn) >
(m/2)Q. Then, Yn(t n ) = wn(t n ), and it follows from Lemma 3.5.1 that
+ cn )-(Q+1)
[(tn - t + cn )-(Q+1) + (m/2)Q+1]
a(tn - t
y~(t)
<
a [(tn - t+ cn)-Q
- a(m/2)Q+1
+ (m/2)Q] (Q+1)/Q
aIYn(t)I(Q+l)/O' - a(m/2)Q+1
on
- a(m/2)Q+1
h.
Thus, we have
+ a(m/2)Q+1 < 0
- alw n (t)I(Q+1)/Q + a(m/2)0'+1
- aIYn(t)I(Q+1)/a
y~(t)
w~(t)
for all t E 12 = In h c [t*, (0). Now, a comparison argument gives

Yn(t) :S Xn(t) on h. Hence, it follows from
wn(t n )
that tn
+ Cn E I
w(t n ) - 2a(m/2)a+ltn > a (1- 2-0') mc<+ltn

for all n large enough.
Oscillation and non oscillation of half-linear diHerential equations
111
By the definition of Yn(t) we find that for all sufficiently large n,

Yn(t) = 00, and hence
limt-4(tn+cn)-
lim
t-4(t n +C n )-
wn(t) = 00.
(3.5.17)
Now, we take k large enough so that tk + Ck E Ik = [tk' (2,,+1 - l)tk].

Clearly, there exists a positive constant M such that
Finally, it follows from (3.5.13) and (3.5.17) that
which is a contradiction. This completes the proof.
Theorem 3.5.2. If q(t) E n satisfies (3.5.3) and M{lql} > 0, then

equation (3.5.1) is oscillatory at + 00 and - 00 for every A E lR - {O}.
Proof. Without loss of generality we shall only show that equation (3.5.4)
is oscillatory at + 00. Assume to the contrary that (3.5.4) has a solution
x(t) which is nonoscillatory at +00. Thus, we can assume that there exists
to> 0 such that x(t) > 0 on [to, 00). Define w(t) = -1jJ(x'(t))N(x(t))
for t 2 to. Then, w(t) is a solution of (3.5.5) on [to, 00). Hence, for any
fixed 6> 0, we have
~ It+li q(s)ds = _
w(t: 6)
+ w~t) +
JIt+li Iw(s)I(,,+1)/"ds
on [to, 00). Applying the Besicovitch semi-norm II . liB"

restriction of I . liB to the interval [to, 00) defined by
IlflIB'
lit
limsup t-400
to
(3.5.18)
essentially a
If(s)lds
to (3.5.18), we find for all 6> 0,
(3.5.19)
Chapter 3
112
It follows from Lemma 3.5.2 that M {Iwl(<>+l)/e>} = o. Thus, IIwll8' =

IIw(t + 8)118' for all 8> o. Using Fubini's theorem, we obtain
~
8t
it lsHIW(~)I(e>+l)/ad~ds
~ it f'5Iw(~ + s)l(a+l)/ad~ds ~ i Iw(~ + s)l(a+l)/adsd~
io
io
11
i
to
at
< -1
8t
tH Iw(s)l(a+l)/Qdsd~ =
-1
to
{6
8t
to
to
tO
to
+6 Iw(s)l(a+l)/ads,
(3.5.20)
for any fixed 8 > O. Now, from (3.5.20) and Lemma 3.5.2, we get
(3.5.21 )
Applying (3.5.21) and JJWIJB'
= Ilw(t + 8)IJB' = 0 to (3.5.20), we find
II~ itH q(S)dst,
= 0 for all
<5 >
o.
(3.5.22)
Finally, since q(t) is Besicovitch almost periodic, it follows from [9J that
This and (3.6.22) imply that M{JqJ}
tion. This completes the proof.
JlqJIB' = 0, which is a contradic-
Example 3.5.1. Consider the half-linear differential equation
(~(X'(t)))' - Acost ~(x(t)) = 0,
t 2: to 2: O.
(3.5.23)
Then, q(t) = cos t, and hence
M{q} =
and
M{JqJ} =
lim -1
t-+oo
1t
0
11t
lim -
t--+oo
Jq(s)Jds
q(s)ds
sin t
.
1I
mt
t-+oo
1
lim 2 ( )
t-+oo
n + 1 7r
2 (n+l)7f
217f/2
lim I cos slds =
n-+oo 7r
I cossJds
2
7r
> O.
113
Thus, it follows from Theorem 3.5.2 that for each A E IR - {O} equation
(3.5.23) is oscillatory at + 00 and - 00.
3.6. On Zeros of Solutions of a Singular

Half-linear Differential Equation
In this section, we shall consider the differential equation
x"(t)
where a E (0,1]
integrable and
q(t)lx(t)IOlx'(t)l l -
sgn x(t),
(3.6.1)
is a constant, the function q: (a, b) -+ IR is locally

(3.6.2)
Sufficient conditions for the existence of a solution of (3.6.1) having at least

two zeros in the interval [a, b] are presented.
In what follows we will use the following notation: L:1oc((a, b)) is the set
offunctions q: (a, b) -+ IR which are Lebesgue integrable on each segment
contained in (a, b). AC([a, b]) is the set of functions u: [a, b] -+ IR
absolutely continuous on the segment [a, b]; ACLoc (I) , where I C IR
is the set of functions u: I -+ IR absolutely continuous on each segment
contained in Ij ACloc(I) , where I C IR is the set of functions u E
AC1oc(I) for which u' E AC1oc(I); u(s+) and u(s-) respectively
are the right and left limits of the function u at the point s; and
q-(t)
(lq(t)l- q(t))/2.
By a solution of equation (3.6.1), where q E L:1oc((a, b)) we mean

a function x E ACloc((al, bl )), where al E [a, b) and bl E (aI, b]
which satisfies (3.6.1) almost everywhere in (aI, bd. In the case of linear
differential equation, i.e., when a = 1, the number of zeros of two arbitrary
non-trivial solutions differ from each other by not more than 1. This fact in
general does not hold for (3.6.1) when ai-I, since any constant function
is also its solution; however, it remains valid for a definite subset of the set
of solutions, which in what follows will be called the set of proper solutions.
Definition 3.6.1. A solution x(t) of (3.6.1) is said to be proper if there
exists A C IR such that mes A = 0 and {t E (a,b) : x'(t) = O} c {t E
(a, b) : q(t) = O} U A.
Definition 3.6.2.
The function q(t) is said to belong to the set
Oo((a, b)) if there exists a proper solution of (3.6.1) having at least two
zeros on the segment [a, b].
Chapter 3
114
In other words, q f:- OQ((a, b)) if and only if there is no proper solution
x(t) of (3.6.1) satisfying for some al E [a, b) the conditions
x(al +) = 0,
x(b l -) = 0.
(3.6.3)
Definition 3.6.3. The function q(t) is said to belong to the set Uc>((a, b))
if for any al E [a,b) and bl E (al,b] the problem (3.6.1), (3.6.3) has no
non-zero (not necessarily proper) solution.
It is clear that if q E Uc>((a, b)), then q f:- Oc>((a, b)). In the case when
a = 1, or a E (0,1) and q(t):s;
for a < t < b, the converse is also
valid, i.e., if q f:- Uc>((a,b)), then q E Oc>((a,b)).
For the convenience of reference we state the following proposition.

Proposition 3.6.1. The following equalities hold
lim (t-a)C>
t--+a+
(a+b)/2
Iq(s)lds =
and
lim (b-t)c>
it
t--+b-
Lemma 3.6.1. There exist two solutions Xl (t) and

(3.6.1) satisfying respectively the initial conditions
xI(a+) = 0,
x2(b-) = 0,
x~ (a+)
x~(b-) =
(a+b)/2
X2 (t)
Iq(s)lds = 0.
of the equation
(3.6.4)
- 1.
(3.6.5)
Moreover, all noncontinuable solutions of the problems (3.6.1), (3.6.4) and

(3.6.1), (3.6.5) are defined on the entire segment (a,b).
Proof. We will prove only the existence of Xl' The existence of X2
can be proved similarly. Let T E (a, b). Denote by nT the set of
all noncontinuable to the right solutions of (3.6.1) satisfying the initial
conditions
x(T) = 0, x' (T) = 1.
(3.6.6)
Let x(, T) E T . We shall show that this solution is defined in the interval
[T, b). Suppose that
Ir = {tE(T,b): Ix(s,T)I<oo, Ix'(s,T)I<oo for T:S;s:s;t}.

Integrating (3.6.1) and taking into account (3.6.6), we obtain
x'(t, T)
1 + l:q(S)IX(S, TWlx'(s, TW-c> sgn x(s, T)ds for t E IT

(3.6.7)
and
x(t, T)
(t-T)+
It
(t-s)lx(s, TWlx'(s, TW-c> sgn x(s, T)ds for t E IT.

(3.6.8)
115
Ix'(t,T)1 ::; 1+ h(s-a)O<lq(s)1
(IXs(~iIO<)
Ix'(s,TW-O<ds for tEIr,
1~(~ i 1 ::; 1+ it (s-a)O<lq(s)1 (I :(~ i 10<) Ix/(s, T)II-O<ds for t
Jr.
Adding the above inequalities and using the fact that

yO<
< 1 + y for
> 0,
(3.6.9)
we find for t E IT,
Applying Lemma 1.1.1, we get
1~(~il+lx'(t,T)I::;
Consequently, sup IT
obtain
2exP (2h(s-a)0<Iq(s)ldS)
for
tEIT.
(3.6.10)
b. Moreover, from (3.6.7) and (3.6.10), we also
Ix(t, T)I < 2(t - a) exp (2it (s - a)O< Iq(S)ldS)

t
Ix'(t,T)I::; 2exP (2i (s-a)0<Iq(8)lds)
for T::; t < b, (3.6.11)
for
T::;t<b
(3.6.12)
and for T::; t < b,
Ix/(t, T) -
11 ::;
2 (it (s - a)O<lq(s)ldS) exp (2it (8 - a)O<lq(s)ldS) .
(3.6.13)
Let Tk E (a, b), Tk+l < Tk for k = 1,2,, limk-4oo Tk = a. Suppose
Uk(t) = x(t, n) for Tk ::; t < b for k = 1,2,. From (3.6.11) and
(3.6.12) the sequences {udk=l and {u~}k=l are uniformly bounded
and equicontinuous in (a, b), (i.e., on each segment contained in (a, b)).
Hence, without loss of generality, by the Arzela-Ascoli theorem, we can
assumethat limk-4oou~i)(t)=x~i)(t), i=O,1 uniformlyin (a,b).
It is easy to see that the function XI(t) is a solution of (3.6.1). Further,
by (3.6.11) and (3.6.13) the function Xl (t) satisfies condition (3.6.4) also.
Thus, we have proved that the set of solutions n of the problem (3.6.1),
(3.6.4) is non-empty.
Chapter 3
116
Now, we shall show that all functions in n are defined on the entire
segment [a, b]. For this, let Xl E n be a noncontinuable solution. Then
analogous to what has been done above, we find
l+_l_It (s-a)O'q(s) I Xl(S) 1

t-a a
s-a
Ix~(s)ll-c> sgn xl(s)ds
for tEl,
(3.6.14)
where
I = {t
(a, b): IXl(S)1 <
Ix~(s)1
00,
<
00
for
a <:: s <:: t}.
These two inequalities on I immediately yield
(b-t)lx~(t)1
<:: (b-a)+
IXl(t)1 <:: 1 +
t-a
~b
1
-a
It
a
It
(s-a)C>(b-s)CXlq(s)
:l~S~ Ie> l(b-s)x'(s)ll-"ds,
(8 _ a)"(b _ s)"lq(8)11 Xl(S) 10' I(b s-a
s)x~ (s)11-ads.
Adding these inequalities, using (3.6.9) and applying Lemma 1.1.1, we obtain
IXl(t)1 +(b-t)lx'(t)1
t- a
<
(1
+b-
Ib
a) exp (
l+b-a
b_ a
(s - a)"(b - s)a1q(s)ldS)
for tEl.
The inequality (3.6.14) for a < t < b results in
)I
(3.6.15)
a
[(s-a)(b-sW q ( S )IXl(s)l
-b-s
s-a
x I(b - S)X~ (8)1 1 - " sgn xl(s)ds
Xl ()
t -_ t - a - (b - t
[(s--a)(b-s)]oq(S)
I:l~; 10' l(b-s)x~(sW-"sgn
xl(s)ds
from which in view of (3.6.15) and Proposition 3.6.1, we conclude that there
exists a finite limit Xl (b-).
The proof of the following lemma is similar to that of Lemma 3.6.l.

Lemma 3.6.2. All noncontinuable solutions of (3.6.1) are defined on the
entire segment [a, b].
117
Remark 3.6.1. If c E (a, b) and x(t) is a solution of (3.6.1) satisfying

the conditions x(c) = 0, x'(c) = 0 then x(t) is identically zero.
Lemma 3.6.3. If x(t) is a solution of equation (3.6.1), then
lim (t - a)lx'(t) I =
t--+a+
0 ( lim (b - t)lx'(t)1 0).
(3.6.16)
(3.6.17)
t--+b-
Proof. We claim that
liminf (t-a)lx'(t)1
~~
(liminf (b-t)lx'(t)1
t~-
0).
For this, assume that (3.6.17) does not hold. Then there exist c
and E > 0 such that
t~a
Ix'(t)l>
for
a<t<c
(lx'(t)l>
b~t
for
(a, b)
C<t<b).
An integration of this inequality from t to c (from c to t) yields
c-a
Ix(t) - x(c)1 > d n - t-a
(Ix(t)-X(C)I >
a<t<c
for
dn~=~
for
c<t<b),
which is impossible because the function x(t) is bounded. Thus, (3.6.17)

holds.
Multiplying both sides of (3.6.1) by (t-a) (by (b-t)) and integrating
it from T to t (from t to T), we obtain
(t - a)x'(t) = (T - a)x'(T)
it
-iT
+
(
+ x(t) - X(T)
(s - a)q(s)lx(s)IOlx'(s)11-.O sgn x(s)ds for a < T < t < b
(b - t)x'(t)
(b - T)X'(T)
+ X(T)
(3.6.18)
- x(t)
(b - s)q(s)lx(s)IOlx'(s)11-O sgn x(s)ds for
a<t<r<b)
from this and (3.6.9), we conclude that
(t - a)lx'(t)1 S (T - a)lx'(T)1
it
(s - a)"'lq(s)I(lx(s)1
+ Ix(t) -
+ (s -
x(T)1
a)lx'(s)l)ds for a < T < t < b

(3.6.19)
Chapter 3
118
(b - t)~'(t)1 :::; (b - 7)lx'(7)1
+ Ix(t)
- x(7)1
(b - sr"lq(s)I(lx(s)1 + (b - s)lx'(s)l)ds for a < t < 7 < b.
Suppose
c(T)
2max{lx(t)-x(a+)I:
for a:::;t:::;T}
+ iT(S - a)'"'lq(s)llx(s)lds
( crT)
2 ma:{lx(t) - x(b-)I:
a<T <b
for
for T:::; t :::; b}
1r (b - sr'lq(s)llx(s)lds
) ,
a<T <b
for
and
M(7, T) = (7 - a)lx'(7)1 + c(T) for a < 7 :::; T < b

(M(7, T) = (b - 7)lx'(7)1 + c(T) for a < T :::; 7 < b).
Then from (3.6.19), we find
(t - a)lx'(t)1 :::; M(7, T) +
(b-t)IX'(t)1 :::; M(7,T)
it
I
(8 - a)"'lq(s)I((s - a)lx'(s)l)ds
for
(3.6.20)
<
<t <T <b
(b-S)''',q(S)'((b-S)'X'(S)')ds).
for
<T <t<7<b
(t-a)lx'(t)1 :::; M(7, T) exp (iT (s - a)"'lq(s)ldS) for a < T < t < T < b
((b-t)IX'(t)1 :::; M(7, T) exp (J:(b-S)"'lq(S)ldS) for a < T < t < 7 < b)
Now, in view of (3.6.17) and (3.6.20), we find
(t - a)lx'(t)1 :'S c(T) exp (iT (s - a)"'lq(s)ldS)
((b-t)IX'(t)l:'S C(T)ex p
(1r (b-S)"'lq(S)ldS)
b
for
for
a<t<T <b
a<T<t<b).
Oscillation and nonoscillation of half-linear diHerential equations

Since limT--+a+ c(T)
that (3.6.16) holds.
=0
(limT--+b~
119
c(T) = 0), the above inequality implies
Lemma 3.6.4. Let al E (a, b), bl E (al' b) and let x(t) and y(t) be
solutions of equation (3.6.1) satisfying the conditions
x(al) = 0, x'(ad > 0 (x(bd = 0, x'(bl ) < 0)

y(t) > 0 for al:::: t :::: bl , y'(bd = 0 (y'(al) = 0).
(3.6.21 )
Furthermore, let u E AC l ([al' bl ]) be such that
u(t) > 0, u'(t) > 0 for al :::: t :::: bl (u(t) > 0, u'(t) < 0 for al :::: t :::: bl )
'ul/(t) :::: q(t)lu(t)I"lx'(t)ll-a for al < t < bl .
Then the following hold
x'(t) > 0 for al < t < bl
(x' (b) < 0 for al :::: t :::: b)
(3.6.22)
and
y'(t)
u'(t)
y(t) < u(t) for al:::: t :::: h
u'(t)
)
( Y'(t)
y(t) > u(t) for al:::: t :::: bl .
(3.6.23)
Proof. Suppose to the contrary that (3.6.22) does not hold. Then by
(3.6.21) there exists c E (al,b l ) such that x'(t) > 0 for a] < t <
c (x'(t) < 0 for c < t :::: bl ) and x'(c) = O. Define
p(t)
u(t) I" sgn u,(t)

Iu'(t)
for
al
< t < bl
and
o-(t) = Ix'(t)
I x(t)
IQ
sgn x'(t)
for
a<
t:::: c
for
al
(for c:::: t
< bd.
Clearly, it follows that
p'(t) :::: aq(t) - alp(t)l(a+I)/a

o-'(t) = aq(t) - alo-(t)l(o+l)/a
for
< t < bl
(3.6.24)
al < t < C (for C < t < bl ) (3.6.25)
o-(al+) = +00, o-(c) = 0 (o-(b l -) = -00, o-(c) = 0).
(3.6.26)
From (3.6.26) there exist Cl E (al'c) and C2 E (CI'C) (Cl E (cl,b l ) and
C2 E (c, cd) such that
p(t) > o-(t) > 0 for Cl < t < C2, P(Cl) = o-(cd
(p(t) < o-(t) < 0 for C2 < t < Cl, P(Cl) = o-(Cl))'
(3.6.27)
120
Chapter 3
Thus from (3.6.24) and (3.6.25), we have
aCt) =
a(Cl)
+ Q:
it
q(s)ds -
Q:
c}
a(t)
Q:
Q:
1~ Ip(s)l(a+l)/ads
2:: pet) for
leI q(s)ds + leI la(s)l(a+l)/ads

leI q(s)ds + leI Ip(s)l(a+l)/ads ~ pet)
= a(cl) -
< P(Cl) -
la(s)l(a+l)/ads
Cl
> P(Cl) + Q: 1~ q(s)ds -
it
Q:
Cl
< t < C2
)
Q:
Q:
for
C2
< t < Cl
'
which contradicts (3.6.27). Hence, (3.6.22) holds. Similarly, we can prove

(3.6.23). This completes the proof.
Lemma 3.6.5. Let x(t)

satisfying the condition
be a nontrivial solution of equation (3.6.1)
x(a+) = 0
(3.6.28)
(x(b-) = 0)
and let u E ACLc((a, cD (u E ACloc([c, b)), c E (a, b) have a finite limit

u( a+) 2:: 0 (u( b-) 2:: 0) and satisfy the conditions
u'(t) > 0 for a < t
(u'(t) < 0 for
ul/(t) ~ q(t)lu(t)lalu'(tw- a for
a < t < C (for
t < b),
C
< t < b).

x' (t)
1=
for
a<t ~ c
(for
~ t
< b).
(3.6.29)
Proof. Let Xl(t) be a solution of the problem (3.6.1), (3.6.4). First we

claim that
(3.6.30)
X~ (t) > 0 for a < t ~ c.
If false, then there exists
Cl,
x~(cd =
o.
Cl E
(a, c) such that
x~ (t)
> 0 for a < t <
By Lemma 3.6.4, we have

X~ (t)
-- <
Xl(t)
Define
u'(t)
u(t)
for
< t < Cl.
yet) = U'(t)Xl(t) - x~(t)u(t) for a < t < Cl,

IU'(t)Xl(tW- a -Ix~(t)u(tw-a
h(t)
for a < t <
U'(t)Xl(t) - x~(t)u(t)
(3.6.31 )
Cl,
g(t) = q(t)h(t)(Xl(t)U(t))Q
for
a < t < Cl.
121
(3.6.32)
Since the function f(x) = (lxI1-Q - 1) /(x - 1) is bounded, there exists

Mo > 0 such that Ih(t)l::; Mo/(u(t)x~ (t))a) for a < t < Cl. From
(3.6.32), we find
< Mo(t - a)"'lq(t)1 Ct ::j~~ (t))
Ig(t)1
< M(t - a)alq(t)1 for a < t < (a + cd/2,

where
M = Mo sup { Ct
_x:j~~ (t)) a :
a<t< a
Cl } .
Thus, the function g(t) is integrable on the segment [a, (a

Since y'(t)::; g(t)y(t) for a < t < (a
y(t)
(_1(a+c
+ cd/2
:2:
y (a
~ Cl )
exp
:2:
y (a
~ Cl )
exp ( _1(a+cl)/2 Ig (S)ldS)
+ cd/2].
it follows that
Il/ 2 9(S)dS)
a + Cl
for a < t ::; - 2 - '
This inequality and (3.6.31) imply that

liminf y(t) > O.
(3.6.33)
t->a+
However, from the proof of Lemma 3.6.3 and condition (3.6.4), we have
liminf y(t) =
liminf ((t-a)u'(t) (x1(t))

t ..... a+
t- a
t ..... a+
-x~(t)u(t))
::; 0,
which contradicts (3.6.33). Thus, (3.6.30) holds.

Now, we shall show that (3.6.29) holds. By Lemma 3.6.4 there exists
Co E (a, c) such that x(t) -=I- 0 for a < t < co. Without loss of generality
assume that
(3.6.34)
x(t) > 0 for a < t < co.
We claim that
x'(t) > 0 for
a < t < Co.
(3.6.35)
If false, then there exists C2 E (a, co) such that X'(C2) = O. By Lemma
3.6.4, we have x'(t)/x(t) < x~(t)/Xl(t) for a < t < C2. Thus, as above it
follows that yi(t) = gl(t)Yl(t) for a < t < C2, where
Yl(t)
X'(t)Xl(t) - x~(t)x(t)
for
a<t <
C2,
122
Chapter 3
gl(t)
h t
1 ()
_
-
q(t)h1(t)(x(t)Xl(t'" for a < t < C2,

IX'(t)Xl(tW-'" -lx~(t)x(tW-'"
X'(t)Xl (t) _ x~ (t)x(t)
for a < t < C2,
=
which as before leads to a contradiction.

Hence, if for some Co E (a, c) inequality (3.6.34) holds, then (3.6.35)
also holds. Therefore, x(t) > 0 for a < t :::; c and (3.6.29) is satisfied.
Remark 3.6.2. Let Xl(t) (X2(t be a solution of the problem (3.6.1),

(3.6.4) ((3.6.5. It is clear that there exists c E (a, b) such that x~ (t) > 0
for a < t < c (x~(t) < 0 for c < t < b). Thus, in view of Lemma 3.6.5 for
the case u(t) = Xl(t), a < t :::; c (u(t) = X2(t), c:::; t < b), if x(t) is a
nontrivial solution of equation (3.6.1) satisfying (3.6.28), then there exists
c E (a, b) such that (3.6.29) holds.
Lemma 3.6.6. Let x(t)
satisfying the conditions
be a nontrivial solution of equation (3.6.1)
liminf (t - a) Ix'((t
x(a+) = 0,
t---+a+
lim inf (b - t) Ix'((t
(X(b-) = 0,
t---+b-
00
< 00).
<
(3.6.36)
x'(t)
X t
.
( hm
(b - t) -x'(t)
Hbx(t)
lim (t - a)-(-) = 1
t---+a+
1) .
(3.6.37)
Proof. By Remark 3.6.2, without loss of generality, we can assume that

for some Co E (a, b),
x'(t) > 0 for
a < t < Co
(x'(t) < 0 for Co < t < b).
(3.6.38)
Multiplying both sides of (3.6.1) by (t - a) (by (b - t and integrating

it from T to t (from t to T), we obtain equality (3.6.18). Using (3.6.9)
and (3.6.38) in (3.6.18), we obtain
(t _ a) x'(t)
x(t)
:::;
()X'(T)
1 + T - a X(T)
(b _ t) Ix'(t)1
x(t)
<
[1
It( S T
+ (s - a) X'x((s)]
s)
ds
)"'1 ( )1
for
qs
a < T < t < Co
l+(b_T)lx'(T)1 + r(b-s)"'lq(s)1
X(T)
it
x
[l+(b-s)l~g}l]dS
for
co<t<T<b
(3.6.39)
Oscillation and nonoscillation of half-linear differentia'] equations

and
I
I(t-a )-x'(t)
--1
x(t)
123
< (T-a)x'(T)+X(T)
rt
x(t)
x' (s) ]
+ JT (s - a)"'lq(s)1 1 + (s - a) x(s) ds for a < T < t <
(b - t)-x'(t)
-+l I
I
x(t)
(
< (b - T)lx'(T)1 + X(T)

x(t)
+ IT(b - s)"'lq(s)1 [1 + (b-s) :~(~n ds for
Co
)
Co
<t <T<b
.
(3.6.40)
From (3.6.39), we find
(t-a)
~g}
<::; M(T)+
(b-t)l~g}1
(s-a t1q (s)1 ((s - a)
<::; M(T)
~(~}) ds
+ I (b-S)'''lq(S)1
for a < T < t < Co
((b-s)I~(~}I)dS),
for Co < t <
where
M(T)
X'(T)
1 + (T - a) X(T)
(M(T) = 1 + (b - T)
jeo(s - a)Qlq(s)lds
I~(~}I + l>b -
s)"lq(s)lds
 0 such

that
x'(t)
(lx'(t)1
)
(t-a) x(t) <::; Mo for a < t < Co
(b-t) x(t) <::; Mo for Co < t < b .
(t-a)x'(t)
x(t)
-11
<::; (T-a)x'(T)+X(T)
x(t)
+ (I+Mo)jt(s-a)Qlq(s)ldS
a
for a
<T <t <
Co
124
Chapter 3
+ 11 :. : : (b-T)lx'(T)I+x(T) + (1+ Mo) Ib(b ( I(b-t) x'(t)

x(t)
x(t)
S)"'lq(S)ldS)
for Co < t <
<b
Thus, by Lemma 3.6.3, we conclude that
l(t-a)~g?-11
)l
)l
:.: : : (I+Mo
(l(b-t)~g?+11 : ;
(I+Mo
Hence, (3.6.37) holds.
(s-a)"'lq(s)ldS
for
a<t<co
(b-S)"'lq(S)ldS
for
Co<t<b).
Lemma 3.6.7. Let x(t) be a nontrivial solution of equation (3.6.1)

satisfying the condition (3.6.28). Then, (3.6.37) holds.
Proof. By Remark 3.6.2, without loss of generality, we can assume that

(3.6.38) holds for some Co E (a, b). Multiplying both sides of (3.6.1) by
(t - a) (by (b - t)), integrating from a to t (from t to b) and taking
Lemma 3.6.3 into account, we obtain
(t - a)x'(t) = x(t)
(b - t)lx'(t)1 = x(t)
it
(s - a)q(s)lx(s)I"'lx'(sW-"'ds
+ Ib(b -
for
a < t < Co
s)q(s)lx(s)I"'lx'(s)11-"'ds for Co < t < b) .

(3.6.41)
(b - t) Ix'((t)) I >
( liminf
t-+bX t
liminf (t - a) x'((t)) > 0

t-+a+
Hence, there exist
> 0 and
x(t) < f(t-a)x'(t) for a < t <
Cl
Cl
(a, co)
(Cl
0).
(co, b)) such that
(x(t) < f(b - t)lx'(t)1 for
Cl
< t < b).

(3.6.42)
Using (3.6.38) and (3.6.42) in (3.6.41), we get
(t-a)x'(t) :.: : : X(T)+f'"
~b-t)IX'(t)1 :.: : : x(T) +
f'"
(s-a)""q(s)'[(s-a),x'(s) '1 ds, a < t:.::::: T <
b
(b-S)"',q(S)'[(b-S)'X'(S) '1 dS,
Cl
Cl
< T :.: : : t < ~ .
125
From this and Lemma 1.1.1, we find

(T _
)x'(T)
a x(T) < exp
((b-T)I:(~?I:::;
(f" I
(s-a)'',q(s)'ds)
eXP(f"lb(b-S)"lq(S)ldS)
for a
for
< T < Cl
Cl<T<b).
Thus, (3.6.36) holds, and hence by Lemma 3.6.6, (3.6.37) follows.
Lemma 3.6.8. Let x(t) be a solution of equation (3.6.1) satisfying the

conditions
x(a+) = 0,
(X(b-)
liminf Ix'(t)1 = 0
t-+a+
= 0,
liminf Ix'(t)1 = 0) .
t-+b-
(3.6.43)
Then, x(t) is identically equal to zero.
Proof. Assume to the contrary that x(t) is a nontrivial solution of

(3.6.1) satisfying the condition (3.6.43). By Remark 3.6.2, without loss of
generality we can assume that (3.6.38) holds for some Co E (a,b). From
equation (3.6.1) for a < T < t < b, we find
x'(t)
X'(T)
it
and
x(t)
X(T)
+ (t -
T)X'(T)
q(s)lx(s)I Ct lx'(s)1 1 -Ct sgn x(s)ds
it
Thus, in view of (3.6.9) for a <
Ix'(t)1 :::; IX'(T)I

Ix(t)1 :::; IX(T)I
t-a
T-a
it
(t - s)q(s)lx(s)I"lx'(s)ll-" sgn x(s)ds.
< t < b, we get
(s - a)"lq(s)1
+ IX'(T)I +
it
T
[~~~ + IX'(s)l] ds,
(s _ a)"lq(s)1 [lx(s)1
s-a
+ IX'(s)l] ds.
Hence, by Lemma 3.6.7 there exist f > 0 and Cl E (a, co) so that
(3.6.42) holds. Combining the last two inequalities, using Lemma 1.1.1 and
condition (3.6.42), we obtain for a < T < t < b, T < CI,
Chapter 3
126
Now, from (3.6.43) and the above inequality it follows that Ix(t)1 + (ta)lx'(t)1 = 0 for a < t < b, which is a contradiction. This completes the
proof.
Lemma 3.6.9. The set of proper solutions of equation (3.6.1) is nonempty.

Proof. Let al E [a,b) and bl E (alobJ. Denote by B([alob}j) the set
of all nontrivial solutions of equation (3.6.1) satisfying the condition
mes{{t E (a},b}): x'(t)
= O} \ {t
(a},b}): q(t) = O}} = O.
i- 0.
We will show that B([a, bj)
First we shall show that if Xo E B([a}, b}j), where b} < b then there
exist b2 E (b},b) and a nontrivial solution Xo of (3.6.1) such that
(3.6.44)
Indeed, by Remark 3.6.1, either
xo(b)
or
xo(b) = 0,
i-
(3.6.45)
0,
x~(b)
i- o.
(3.6.46)
Suppose (3.6.45) holds. Denote by Po the solution of the Cauchy problem
PI (t) = aq(t) - alp(t)1 ( '" +})/ "',
p(b}) =
1x~(b})
I
xo(b}) I'" sgn xo(b}).
Assume Po(t) is defined on [b},b 2 J. Let
x(t) _
-
{XO(t)
for
xo(b l ) exp
(la\::; t ::;
b
b}
Ipo(sW/'" sgn Po(s)ds
for
bl
::;
t ::; b2 .
The function x(t) clearly satisfies equation (3.6.1) for t E (a, ~). By
Lemma 3.6.2 there exists a solution xo(t) of (3.6.1) such that xo(t) = x(t)
for t E [a}, b2 J and satisfies (3.6.44).
Suppose now that (3.6.46) is satisfied. Denote by Yo(t) a solution of
the Cauchy problem
x"(t) = q(t)lx(t)I"'lx'(t)II-'" sgn x(t),
x(b 1 ) = 0, x'(b}) = x~(bd.
Obviously, there exists b2 E (b l , b) such that y~(t) i- 0 for t E [b l , b2 J.

Suppose
for a}::; t ::; b1
x(t) = { xo(t)
yo(t) for
b}::; t ::; b2 .
Oscillation and non oscillation of baH-linear differential equations
127
Again, it follows that there exists a solution xo(t) of (3.6.1) satisfying

(3.6.44) .
Now, let Xl(t) be a solution of the problem (3.6.1), (3.6.4). It is easy
to see that there exists 0,1 E (a,b) such that Xl E B([a,al]). Denote by
I the set of those bl E [0,1, b] for which there exists a solution x( t) of the
equation (3.6.1) such that x(t) = Xl(t) for a:<::; t :<::; 0,1, X E B([a, b]). Let
bl = sup I. By Lemma 3.6.2, to prove the result it suffices to show that
bl = b. Assume to the contrary that bl < b. Then as is shown above if
bl E I, there exists b2 E (bl,b) such that ~ E I. Consequently, Ii- 0,
and I is a connected set whose every point, except the point 0,1 is an
interior point. In view of this fact, we have
b
rf.
(3.6.47)
I.
Choose a sequence of points {tdk:,l and a sequence of functions {Yk(t) }k=l

such that tk < tk+l, k = 1,2,, 0,1 < tl, limk->oo tk = bl
Yk
B([a, tk]), k = 1,2, ...

Yl (t) = Xl (t)
for a < t < 0,1
Yk+l(t) = Ydt) for a < t < tk,
E
(3.6.48)
k = 1,2,.
Define
x(t)
Yl ( t)
a:<::; t < iI
for tk-l:<::;t<tk,
for
Yk(t)
k=2,3,.
(3.6.49)
It can be seen that X E ACloc((a,bd) and x(t) satisfies (3.6.1) for

t E (a, ( 1 ). Hence by Lemma 3.6.2 there exists a solution x(t) of (3.6.1)
such that x(t) = x(t) for a:<::; t:<::; bl . From (3.6.48) and (3.6.49), we find
{t
(a, tk) : y~(t) = O}
mes Ak = 0,
{t
{t
(a, tk) : q(t) = O} U A k,
k = 1,2,
(a,bl ): (x(t' = O} c {t
(a,bl): q(t) = O} uA,
Uk=l A k
Since mes A '=' I:~l mes Ak = 0, this implies that x E B([a, bl]), and
consequently bl E I, which contradicts (3.6.47).
Remark 3.6.3. In the above lemma, we proved that the problem (3.6.1),
(3.6.4) has at least one proper solution. Similarly, we can show that the
problem (3.6.1), (3.6.5) has at least one proper solution, and for any to E
( a,b) and Cl, C2 E lR, ICII + 1c21 i= 0 the Cauchy problem
xl/(t)
q(t)lx(t)IQlx'(tW-Q sgn x(t)
x(to)
Cl,
x'(to) =
C2
Chapter 3
128
has at least one proper solution.
Lemma 3.6.10. Let there exist c E (a, b)
y E ACloJ(a, c) U (c, b)) having finite limit
y(a+)
~ 0,
y(b-)
~ 0,
and a continuous function
y'(c-) > 0,
y'(c+) < 0
and satisfying the conditions
y"(t) :S q(t)ly(t)IOly'(tW-a
for
a < t < b,
y(t) > 0 for a < t < b, y'(t) > 0 for

y'(t) < 0 for c < t < b.
a < t < c,
Then, q E Uo((a,b)).
Assume to the contrary. Let q rf. Ua((a,b)). Then there exist
and a solution xo(t) of (3.6.1) satisfying the
conditions
Proof.
al E [a,b), bl E (al,b]
(3.6.50)
By Lemma 3.6.8 and Remark 3.6.1 there exist
h, bd such that
x~(t)
x~(t)
> 0 for
< 0 for
al < t < 'I,

' I < t < bl ,
'I
(al,b l ) and
'2
X~(Tl) = 0
X~(T2) = O.
Let Xl(t) and X2(t) be some solutions of problems (3.6.1), (3.6.4)

and (3.6.1), (3.6.5) respectively. Then by Lemmas 3.6.4 and 3.6.5 (with
y(t) = Xl(t), x(t) = xo(t) and y(t) = X2(t), x(t) = xo(t), respectively),
there exist tl E (a,Tl] and t2 E [iz,b) such that
x~ (t)
x~(t)
> 0 for a < t < t 1 ,

< 0 for t2 < t < b,
x~ (t d = 0
X~(t2) = O.
By Lemma 3.6.4 (with x(t) = Xl(t) and x(t) = X2(t), respectively), we

have tl > c and t2 < c which is impossible, since t2 ~ tl'
Lemma 3.6.11.
Let there exist c E (a, b) and a function 0" E
ACLoc((a, c) U (c,b)) having finite limits O"(c-) ~ dc+) and satisfying
the conditions
O"'(t) :S aq(t) - aIO"(t)l(o+l)/o

lirninf (t - a)O"(t) < 1,
t-+a+
Then, q
rf. O",((a, b)).
for
a<t -l.

t-+b-
(3.6.51)
(3.6.52)
129
Proof. Assume to the contrary that q E Oo((a, b)). Then there exist
al E [a, b), b1 E (al,bJ and a proper solution xo(t) of (3.6.1) satisfying
the condition (3.6.50). Suppose al = a and b1 = b (the lemma for the
other cases can be proved analogously). First, we introduce the function
p by
p(t)
~~~!~
0
1
sgn
x~(t)
for
a < t < b.
Clearly, we have
p'(t) = aq(t) - alp(t)l(o+l)/Ct
for
a<t<b
(3.6.53)
and by Lemma 3.6.6,

lim (t - a)Op(t) = 1,
t-+a+
lim (b - t)Op(t) =
t-+b-
- l.
(3.6.54)
From (3.6.52) and (3.6.54) we conclude that there exist tl E (a, b) and
a E AC([tl' tl + ED, and
E E (0, b - t 1 ) such that

a(t) > p(t)
for
tl < t < tl
+ E, a(td = p(td.
::::; t ::::; tl + E. Then
Suppose u(t) = a(t) - p(t) for tl

integrable function h: (tl' tl + E) -+ IR such that
la(t)I(O+J)/O _lp(t)l(o+l)/o
(3.6.55)
there exists an
= (a(t) - p(t))h(t) for tl < t < tl + E.
It follows from this fact, (3.6.51) and (3.6.53) that u'(t)::::; -h(t)u(t) for
tl < t < tl + E, and hence we conclude that u(t) < 0 for tl < t < tl + E,
Before we present the main results of this section, we introduce the

notation
_
Q(t, to, a) -
{ a(t - a)O
ito
q(s)ds
for
a < t < to
a(b - t)O
q(s)ds
for
to < t < b,
to
Q*(to,a)
Q*(to, a)
inf{Q(t,to,a): a < t < b},

sup{Q(t, to, a) : a < t < b}.
Theorem 3.6.1. Let there exist ,x E (a, b) and I-l E (,x, b) such that
-a
11-'q(s)ds ~
(3.6.56)
Chapter 3
130
Then, q E Oo((a, b)).

Proof. Let xo(t) be a proper solution of (3.6.1). We claim that xo(t)
has at least one zero in the interval (a, b). Assume to the contrary that
xo(t) > 0 for
a < t < b.
for
(3.6.57)
Let
pet)
xo(t)
xo(t)
sgn xo(t)
a<t <b
(3.6.58)
so that
p'(t) = o:q(t) - o:lp(t)l(o+l)/o
for
a < t < b.
(3.6.59)

{t E (a,
Al : (t -
a)O pet)
I} U {t E [tt, b) : (b - t)O pet)

U {t E [A, ttl: pet) = O}
-I}
i= (a, b).
(3.6.60)
Multiplying both sides of (3.6.59) by (t - a)o+1, integrating from a to

A and taking into account Lemmas 3.6.3 and 3.6.7, we obtain
+ A _1 a
Jar>- (s -
A _0: a
Jar>-
[(0: + l)(s - a)<>p(s) - o:(s - a)o+llp(s)I(<>+l)/<>] ds.
(A - a)Op(A)
a)o+lq(s)ds
(3.6.61)
Multiplying both sides of (3.6.59) by (b - t)<>+l, integrating from tt to
b and taking into account Lemmas 3.6.3 and 3.6.7, we get
Finally, integrating (3.6.59) from A to tt, we find

(3.6.63)
Clearly, we have
(0: + l)T - o:ITI(<>+1)/<> < 1 for T E IR.\{1}

(o:+l)T+o:ITI(<>+l)/<> > -1 for TEIR.\{-l}.
(3.6.64)
131
Now, in view of (3.6.60) and (3.6.63), from (3.6.61) - (3.6.63) we obtain

the following three inequalities
(A - a)a peA)
1+ A
~ a 1>' (s -
a)a+lq(s)ds,
(3.6.65)
(b_p,)O<p(p,) > _1 _ _
a_ (b(b_s)o<+lq(s)ds
(3.6.66)
b - p, il-'
-a
11-' q(s)ds
peA) - pep,)
(3.6.67)
of which at least one is strictly satisfied. Using (3.6.65) and (3.6.66) in

(3.6.67), we find that condition (3.6.56) is violated. Thus, every proper
solution of (3.6.1) has at least one zero in the interval (a, b). In particular,
the proper solution of the problem (3.6.1), (3.6.4) (see Remark 3.6.3) has
also at least one zero in the interval (a, b), and hence q E Oa((a, b)).
-(a+l)1bQ (s,a;b,a)ds > b-a.
(3.6.68)
Then, q E O",((a,b)).
Proof.
[(a
p, E
From condition (3.6.68) one can choose A E (a, (a

such that
+ b)/2, b)
and
~)"'+l 1>' Q (s, a; b, a) ds
-a(a + 1) [(A _
+(b_:)o<+l
+ b)/2]
Q(s, a;b,a)ds]
2::
(A_~)O<+l + (b_:)o<+l
(3.6.69)
By Proposition 3.6.1, we can easily verify that
1
a
>.
a +b )
Q ( s, -2-,a
ds
--
+1
[ (A - a)"'+l 1(a+b)/2 q(s)ds

>.
1>' (s - a)a+lq(s)ds]
a
and
{b
il-'
a +b
Q s, -2-,a ds
=
_1_ [(b_P,)o<+l
q(s)ds+ {b(b_S)O<+lq(S)dS].
a +1
i(a+b)12
il-'
132
Chapter 3
In view of these facts, we find from (3.6.69) that the inequality (3.6.56) is
satisfied. Hence, by Theorem 3.6.1, q E Oa((a, b)).
Corollary 3.6.2. Let to E (a, b),
a<t <b
q(t) :S 0 for
a)
to -Q*(to,a) :S - 1 - max { ( b - to
'
b- to )
to - a
(3.6.70)
a }
(3.6.71)
Then, q E Oa((a,b)).
Proof. By (3.6.71) there exists c E (a, b) \ {to} such that
-Q(c,to,a) 2
l+max{(~o_-t:)a, (:o-_t~r'}.
(3.6.72)
Suppose
). =
{ to if c < to
c if c > to.
c if c < to
to if c > to,
From (3.6.72), we conclude that
This and inequality (3.6.70) show that (3.6.56) holds, and hence by Theorem 3.6.1, q E Oa((a,b)).
In the case when condition (3.6.70) holds Theorem 3.6.1 can be improved. In fact, the following result holds.
Theorem 3.6.2. Let condition (3.6.70) hold and there exist ). E (a, b),
j.l E ()., b) and n E IN such that
(3.6.73)
where
al(t)
t - a, ak+1(t) = t - a +
it
ak'(s)lq(s)lds, k 2 1 for a < t < b,
b1(t)=b-t, bk+l(t)=b-t+ l\k'(S),q(S)'dS, k21 for a<t<b.

Then, q E Oa((a,b)).
133
Proof. Let XI(t) be a proper solution of the problem (3.6.1), (3.6.4).

We claim that Xl (t) has at least one zero in the interval (a, b). Assume
the contrary. Then by (3.6.70) either
> 0 for a < t 0, a < t <
x~(t)
CI,
C2
< 0,
(3.6.74)
[cI,b) such that

C2
< t < b,
x~(t) = 0,
~ t ~
CI
C2.
(3.6.75)
Suppose
pet) =
IX~(t)la sgnx~(t)
Xl (t)
for
a<t<b
so that (3.6.59) is satisfied.

Assume that (3.6.75) holds. Multiplying both sides of (3.6.59) by (t-
a)",+l and integrating from a to t, we obtain for a < t < b,
which in view of (3.6.64) results in pet)

now that for some k E 1N,
pet)
ak"'(t)
for
a1"'(t) for a < t < b. Assume

a < t < b.
(3.6.76)
From (3.6.59) with regard to (3.6.76), we have for a < t <
CI,
Integrating the above inequality from a to t, we arrive at
p-l/"'(t)
t -
a+
lot a~+I(s)lq(s)lds
for
a<t <
Cl'
From (3.6.75), we find that pet) ~ ak_;\ (t) for a < t < b. Thus, it is
proved by induction that for any k E 1N the inequality (3.6.76) holds.
Analogously, we can show that for any k E 1N,
pet)
- bk"'(t)
for
a < t < b.
(3.6.77)
Suppose now that (3.6.74) is satisfied. Then, as above, we find that

(3.6.76) holds and the inequality (3.6.77) is trivial in this case.
Chapter 3
134
Integrating (3.6.59) from A to 11. and using (3.6.76) and (3.6.77) (for
Iq(s)lds a;:;-a(A) +b;:;-a(p), which contradicts
(3.6.73).
k = n), we obtain - a
s:
1:
Theorem 3.6.3.
Let there exist c E (a, b)
AC([a, c]) and 9 E AC([c, b]) such that
f(a) = 0,
g(b) = 0,
g(t) > 0 for
a < t < c,
f(t) > 0 for
c<t<b,
and the functions
Ig'(tW+ 1
g<>(t)
and
be integrable on [a, c] and [c, b] respectively, and

-a [g(C)
1
c
f(s)q(s)ds
+ f(c)
> (a + 1)-(<>+1) [g(c)
1
C
lb
g(S)Q(S)dS]
11'(s)la+1 ds
f<>(s)
+ f(c)
Ib
c
Ig'(s)I<>+l dS].
g"(s)
(3.6.78)
Then, q E O,,((a,b)).
Proof. Let xo(t) be a proper solution of (3.6.1). We claim that xo(t)

has at least one zero in the interval (a, b). Assume to the contrary that
(3.6.57) holds. Define the function p( t) by (3.6.58) and obtain (3.6.59).
We can easily verify that for a < t < c,
(a
1t
+ 1)f1/(a+1l(t) <
-
11'(s)1 ds < (t _ a)Q+1

f a/("+1)(S)
-
llt
a
11'(s)I"+l dS]
f"(s)
Q+1
and for c < t < b,
(a+1)gl/("+1)(t) < rb Ig'(s)1 ds < (b-t)Q+1 [rb 19'(s)I"+l dS]

- it g,,/(a+1)(s) it ga(s)
0+'
From Lemmas 3.6.3 and 3.6.7, we have

lim f(t)p(t) = 0,
t--+a+
lim g(t)p(t) = O.
t--+b-
(3.6.79)
Multiplying both sides of (3.6.59) by f(t) and integrating from a + E to

c, where E E (0, c - a), we obtain
-a
1
c
f(s)q(s)ds
+ f(c)p(c)
- f(a
+ E)p(a + c)
a+
1
c
[f'(s)p(s) - a f (s)lp(s)l(a+1)/a] ds
a+
<
a+
f(s) [I1'(S)llp(S)I_ a IP (s)I(<>+l)/a] ds.

f(s)
(3.6.80)
135
Similarly, multiplying both sides of (3.6.59) by g(t) and integrating from

c to b - "1, where "1 E (O,b - c), we get
-0
b-ry
: :; l
g(s)q(s)ds + g(b - "I)p(b - "1) - g(c)p(c)

(3.6.81 )
b- c g(s) [lg'(S)IIP(s)l_ olp(s)l(o+l)/O] ds.

c
g(s)
It can be easily verified that
( _A )"'+1
0+1
for
Thus, we can conclude that
-og(c)
and
-of (c)
f(s)q(s)ds
+ g(c)f(c)p(c) <
g(s)q(s)ds - g(c)f(c)p(c) <

-
g(c)
(0 + 1)"'+1
f(c)
(0
+ 1)"'+1
T E JR.
Ib
c
If'(s)I",+l
ds
f"(s)
19'(s)I"'+1
ds.
go(S)
Addition of the last two inequalities gives a contradiction to (3.6.78).
Corollary 3.6.3. Let one of the following conditions be satisfied
-0
_0
(s-a)o+l(b-s)o+1 q(s)ds>
[1(a+b)/2 (s _ a)Aq(s)ds
a
+ {b
(b
a)0+2
,
0+2
(3.6.82)
(b - S)Aq(S)dS]
J(a+b)/2
2A o +1
(b_a)A-O
> (A _ 0)(0 + 1)0+1 -2,
(3.6.83)
where
A> 0
or
_lbsino+1(7r~s~:))q(S)dS >
2(0+2)'" ( 7r
(0+1)0+1 b-a
)0
(3.6.84)
Then, q E O",a,b)).
Proof. Condition (3.6.82) follows from (3.6.81) in the case when f(t) =
[(t - a)(b - t)]o+1 = g(t) for a < t < b; condition (3.6.83) is obtained in
the case when f(t) = (t - a)A and g(t) = (b - t)A for a < t < b, and
condition (3.6.84) in the case when f(t) = sin (7r(t - a)/b - a) = g(t) for
a < t < b.
Chapter 3
136
Finally, we shall present results which ensure that the function q(t) E
Uo((a, b, and when q(t) tf- Oo.((a, b.
Theorem 3.6.4. Let sup{Ao.(t)!B(t)!I-o.:a<t<b} <b-a, where
A(t)
B(t) =
It
it
(b - t)l-A
(s _ a)l+o.A(b _ s)C>A[q(s)rds
(t-a)A a
a)l-A
(t
+ (;_ t)A t (s - a)O>'(b - s)l+o.>'[q(s)rds,
Ib
(s-a)l+o.>'(b-S)C>A[q(S)l-dS-l (s-a)cx>'(b-S)l+o.>'[q(S)l-ds
and>" E [0,1]. Then, q E U",((a,b.

Proof. Suppose [q(t)t =I' 0, since otherwise q E Uo.((a, b. Choose
E >
such that sup {(A(t) + E)O!B(t)!l-o. : a < t < b} 

for a < t < b, yo(a+) = 0, yo(b-) = E, and
there exists Co E (a, b) such that yb(t) sgn (co - t) >
for a < t , .:::: q(t)!yo(t)!o.!yb(t)!l-c> for a < t < b.
(3.6.86)
Let Xl (t) and X2(t) be proper solutions of the problems (3.6.1), (3.6.4)
and (3.6.1), (3.6.5) respectively. By Lemma 3.6.5 (in this case u(t) = yo(t)
for a < t < b) there exist tl E [co, b) and t2 E (a, co] such that
X~(t)
x~(t)
Suppose tl
>
<
for
for
a<t<tl,
t2 < t < b,
x~(tr)=O
X~(t2) =
0.
(3.6.87)
= t2' Then by (3.6.87) the function

x(t)
Xl (t)
for
Xl (co)
X2(t)-(-)
X2 Co
a < t < Co
Co < t < b
for
will be a proper solution of equation (3.6.1), and hence

q E
Define
l7(t)
~~~~~ 10.
Oo((a, b.
sgn yb(t)
for
(3.6.88)
a <t t2. Then from (3.6.87) there exists c E (tl' t2)
such that x~ (c) > 0 > x~ (c). It can be easily verified that the function
= {
u(t)
Xl(t)
for
Xl(C)
X2(t) X2(C)
a<tS;c
for
<t <b
satisfies the conditions of Lemma 3.6.10 and hence q E Uoa,b)).

C(l-o)/o
where
C = max
{l
(S - a)Ct(b - s)O[q(s)]-ds < (b - a)l/o,
(s-a)HO(b-S)Ct[q(S)]-dS'
(s-a)O(b-s)O+l[q(S)]-ds}.
Then, q E Uo(a, b)).

In particular, this corollary implies that if
(s-a)Ct(b-s)O[q(s)t ds < (b-a)O,
then q E Uoa,b)).
~(A),
(_0_)0<
where A> _ _
1_
0+1 0+1
root of the equation T - ITlo/(o<+l) - A = O.
Denote by
Theorem 3.6.5. Let

- -1-
0+1
( -0 -
0+1
'
the largest positive
)0
and
(3.6.89)
Then, q E Uoa, b)).
Proof. Let k = ~(Q*(to,o)),
aCt)
Q(t, to, 0) - k
for a < t < to
(t - a)o<
{
Q(t, to, 0) - k
(b-t)Ct
for to<t<b.
(3.6.90)
Chapter 3
138
Then from (3.6.89), we have
k - ka/(a+l) :::;; Q(t, to, a) < k
a<t 0 for a < t < b, t =I- to and
la(t)l(a+l)/a :::;; k h(t)
where
h(t) =
( _
t
)a+1
y(t)
exp (sgn (to - t)
(3.6.92)
a < t :::;; to
for
(3.6.93)
(b _ t)a+l
Define
a < t < b,
for
1:
to < t < b.
for
la(s)11/a d s)
for
a < t < b.
From (3.6.90) - (3.6.92) it follows that y E ACloc((a, to) U (to, b,
y'(t) sgn (to - t) > 0 for a < t < b, t =I- to,

y'(to+) > 0 > y'(to-), y(a+) = 0, y(b-) = 0
and
y"(t)
q(t)ly(tWly'(t)1 1 -
+ y(t)la(t)l(l-o)/a
(la(t)l(o+l)/a - kh(t)
:::;; q(t)ly(t)loIY'(tW-O for a < t < b.

Hence, by Lemma 3.6.10, we have q E Uo((a, b.
- -1a+1
( -a-
a+1
)0
_ a )a+1
a+1
Then, q E Ua((a, b.
Theorem 3.6.6. Let
- -1-
a+1
( -a-
a+1
)a
and either
or
Q*(t o, a)
and
Q*(to, a) < 2.
(3.6.95)
139
Then, q 'f- Oa((a,b)).

Proof. Let
k = {f,(Q*(to, a)) for Q*(to, a) =f.

kl for Q*(to, a) = 0,
where kl E ((a/(a + 1))a+1, 1) is chosen in such a way that Q*(to, a) <

kl + k~/(a+1). It follows that k E (0,1).
Define the function u(t) by (3.6.90). Clearly,
AC!oc((a, to)
u(to+)
(to, b)),
-(b-to)a'
u(to-) = (to _ a)a '
lim (t - a)au(t)
t ..... a+
k<1
and limHb- (b - t)au(t) = -k > -1. In view of (3.6.94) and (3.6.95), we

can conclude that
k - k a /(a+1) ::; aQ(t, to, a) ::; k + ka/(a+l)
for
a < t < b.
Thus, (3.6.92), where h(t) is the function defined by (3.6.93) is satisfied.

In view of this fact for a < t < b, we have
u'(t) = aq(t) - akh(t) ::; aq(t) - alu(t)l(a+l)/a.

By Lemma 3.6.11, we now have q E Oa((a, b)).
- -1- ( -a- ) a < Q*(to, a) and Q*(to, a) < 2a + 1 (

a+1
a+1
a+1
a ) a
a+1
Then, q 'f- Oa((a, b)).

We note that Corollaries 3.6.5 and 3.6.6 follow from the fact that
Corollary 3.6.7. Let a = 1, Q*(to, 1) ::::: -1/4 and either
Q*(to, 1) =f. 0,
Q*(to, 1) ::; 1 + Q*(to, 1) +
or
Q*(to, 1) = 0,
Then, q 'f- Ol((a,b)) (i.e"
\/1 + 4Q*(to, 1),
Q*(to, 1) < 2.
q E U1((a, b)).
140
Chapter 3

We begin with the following nonoscillatory characterization of (3.1.1).
3.7.1. Nonoscillatory Characterization

exist a number T 2:': to and a function w(t) E C 1 ([T, 00), lR) such that
+ aa- 1/<>(t)lw(t)l(a+l)/0 + q(t)
w'(t)
[T,oo).
::::: 0 on
(3.7.1)
Proof. Let x(t) be a nonoscillatory solution of (3.1.1). Then there is

a number T 2:': to such that x(t) i= 0 for all t 2:': T. Define w(t) =
a(t)~j;(x'(tN(x(t
on [T,oo). Clearly, w(t) E C1([T,00),lR) and
satisfies (3.7.1). Conversely, let inequality (3.7.1) have a solution w(t) on
[T,oo). Define - q1(t) = w'(t) + aa- 1/0(t)lw(t)l(a+1)/a on [T,oo). It
follows from (3.7.1) that
q1(t) 2:': q(t)
for all
t 2:': T.
(3.7.2)
A simple calculation shows that
u(t)
is a solution of
Isl(1-</os.
exp ( [ 1j;-1
(a(t)1j;(x'(t)'
(:f:j) ) ds
+ qdt)1j;(x(t =
Since u(t) does not vanish on [T,oo),

imply that equation (3.1.1) is nonoscillatory.
0,
where
1j;-l(s) =
(3.7.2) and Corollary 3.2.2
Now, we shall study the nonoscillatory behavior of (3.1.1) under the

conditions
10
00
a-1/O(s)ds
00
and
Joo q(s)ds
< 00.
(3.7.3)
It:
In what follows we shall frequently use the functions R(t) =

a- 1/<>(s)ds
oo
and Q(t) = h q(s)ds.
(i)
(ii)
Equation (3.1.1) is nonoscillatory.

There exists a function w(t) E C([T,oo),lR), T 2:': to such that for
t 2:': T,
(3.7.4)
Oscillation and non oscillation of half~linear differential equations

(iii) There exists a function w(t) E C([T,oo),lR), T
~ T,
141
to such that for
(3.7.5)
Proof. (i) => (ii). Let x(t) be a nonoscillatory solution of (3.1.1).
Without loss of generality, we can assume that x(t) > 0 on [T,oo) for
some T ~ to. Define
1j;(x'(t))
w(t) = a(t) 1j;(x(t))
on
(3.7.6)
[T,oo).
From equation (3.1.1), we find
w'(t)
Let t
+ aa~l/<>(t)lw(t)I(Q+l)/<> + q(t)
on
[T,oo).
T be fixed arbitrarily. Integrating (3.7.7) over
w(~)-w(t)+a
If. a~l/Q(s)lw(s)I(<>+l)/<>ds+If.
[t,~],
(3.7.7)
we get
q(s)ds = 0 for
~ ~ t ~ T.
(3.7.8)
We claim that
(3.7.9)
Suppose to the contrary that ftOO a~l/Q(s)lw(s)I(<>+l)/Qds = 00. Then
in view of (3.7.8) and (3.7.3) there is a To ~ t large enough so that for
~ ~ To,
w(O
=
rf. a~l/<>(s)lw(s)I(<>+l)/<>ds
w(t) -If. q(s)ds - a ITO a~l/Q(s)lw(s)I(Q+l)/Qds
+a
lTo
< - 1,
or equivalently,
-w(~) ~ 1 + a
rf. a~l/<>(s)lw(s)I(Q+l)/Qds
lTo
for
~ To.
(3.7.10)
Thus, it follows that x' (~) < 0 and
(3.7.11)
142
Chapter 3
Integrating (3.7.11) from To to
~,
we obtain
which when combined with (3.7.10) implies that - x'(~) ~ a-l/a(~)x(To)

for ~ ~ To. Integrating the last inequality and using (3.7.3), we find that
x(~) ----+ 00 as ~ ----+ 00, which contradicts the assumption that x(t) > 0
for t ~ T. Therefore, (3.7.9) must hold.
We now let ~ ----+ 00 in (3.7.8). Using (3.7.9) and (3.7.3), we observe
that w(O tends to a finite limit w(oo). But, x(oo) must be zero,
otherwise, in view of (3.7.3), (3.7.9) must fail. Thus, (3.7.4) holds.
(ii)
(iii)
(iii). The proof is clear.

(i). Let
w(t)
00
q(s)ds + a
00
a-1/a(s)lw(s)l(a+l)/ads.
Then, (3.7.1) holds. Now, it follows from Theorem 3.7.1 that equation
To prove our next result, we need the following lemma.

Lemma 3.7.1. The function
h(x) = la + axl(a+l)/a - (a + 1)laI 1/ a x sgn a,

where a is a real number, is nondecreasing on [0,00).
Proof. Since
h'(x)
= {
(a + 1)[a + axp/a - (a + 1)laI 1/ a sgn a ~ 0 if a + ax ~ 0

-(a + 1)[-a - axj1/a + (a + 1)laI1/a ~ 0 if a + ax < 0,
h(x) is nondecreasing on [0,00).
In what follows we shall let by
Theorem 3.7.3. Let condition (3.7.3) hold. Then the following three
statements are equivalent:
(i)
Equation (3.1.1) is nonoscillatory.
143

(ii) There exist a number T
such that
y(t)
00
[IQ(s)
to and a function y(t)
+ O:Y(S)I(o+l)/o -
C([T, 00), lR)
(0: + l)IQ(sW/oY(s) sgn Q(s)]

x
a-1/o(s)J.l[S, t]ds
(3.7.13)
(iii) There exist a number T

such that
Z(t) =
00
to and a function z(t)
+ o:z(s)I(O+l)/a -
[IQ(s)
C([T,oo),lR)
(0: + 1)IQ(s)ll/Qz (s) sgn Q(s)]
x a-1/a(s)J.l[S, t]ds.
(3.7.14)
Proof. (i) =} (ii). Assume that equation (3.1.1) is nonoscillatory. It

follows from Theorem 3.7.2 that (3.7.4) has a solution w(t). Define y(t) =
ftOO a-1/a(s)lw(s)l(a+l)/ads ~ O. Clearly, y'(t) = _a-1/a(t)lw(t)l(a+l)/a.
Using J.l[s, t] as an integrating factor and integrating from t to T (~ t),
we obtain
y(t)
+
~
J.l[T, t]y(T)
I
l
[IQ(S)+O:Y(sW-::! -
0+1
(o:+I)IQ(s)l~y(s) sgn Q(s)] a-~(s)J.l[s,t]ds
IQ(s)l---;:;-a-"(s)J.l[s,t]ds.
(3.7.15)
The left hand side of (3.7.15) is independent of T, and hence as T -t- 00,
(3.7.13) follows.
(ii) =} (iii). Suppose there exists a function y(t) E C([T, 00), lR) which
satisfies (3.7.13). Define
Zo(t)
y(t)
zn(t) = lO[IQ(s)+o:Zn_l(s)I(a+l)/a-(O:+I)IQ(sW/aZn_l(s)sgn Q(s)]

x a-1/a(s)J.l[s, t]ds
for n = 1,2,. Then from Lemma 3.7.1 and (3.7.13), it is easy to

show by induction that 0 ~ Ql(t) ~ zn(t) ~ Zn-l(t) ~ y(t). Thus, the
sequence of functions {zn(t)}~=o has a pointwise limiting function z(t) =
limn .... oo zn(t). Now, it follows from (3.7.13) and the Lebesgue dominated
convergence theorem that z(t) satisfies (3.7.14).
Chapter 3
144
(iii) =} (i). Assume that (3.7.14) has a solution z(t), then an easy
computation shows that w(t) = -[Q(t) + az(t)]ja(t) defines a solution of
(3.7.1). Now, as a consequence of Theorem 3.7.1, we conclude that equation
To establish a nonoscillation characterization of equation (3.l.1) when

E C([to, 00), JR+), and
q(t)
7T(t)
00
a-l/<>(s)ds
and
7T(to) < 00
(3.7.16)
we need the following lemma.

Let A be a constant. Then the function
is nondecreasing for X 2" -(Aj(a + l))n.
Lemma 3.7.2.
alxl(n+l)/Ct
+ AX
g(x)
Proof. Since
g'(x)
(a
+ l)lxl"
sgn x
+ A 2"
0 for
x 2" -(Aj(a
g(x) is nondecreasing for x 2" -(Aj(a + 1))<>.
+ I))",
Lemma 3.7.3. Assume that q(t) E C([to, 00), JR+) and 7T(to) < 00. Let
x(t) be a solution of equation (3.1.1) such that x(t) # 0 for t 2" T 2" to.
Then, x(t) as well as the :unction
<>( ) ()1jJ(X'(t))
7T tat 1jJ(X(t)).
(3.7.17)
are bounded on [T, 00). Furthermore,
7TO(t)a(t) 1jJ(x'(t)) > - 1 for t 2" T

1jJ(x(t))
and
lim sup Jr<>(t)a(t) 1jJ~xtN <:::
1jJ x t
t-'>oo
o.
(3.7.18)
(3.7.19)
Proof. Without loss of generality, we assume that x(t) > 0 for t 2" T 2"
to. Since a(t)1/J(x'(t)) is nonincreasing on [T,oo), we find that x'(t)
is eventually of constant sign, i.e., x' (t) > 0 for t 2" T, or there is a
TI > T such that x'(t) < 0 for t 2" TI and
a1/<>(s)x'(s) <::: al/<>(t)x'(t)
for
s 2" t 2" T.
(3.7.20)
Dividing both sides of (3.7.20) by al/<> (s) and integrating it over [t,~]
gives
x(~)
<::: x(t)
+ al/<>(t)x'(t)
if,
a-l/a(s)ds
for
~ 2" t 2" T.
(3.7.21)
Oscillation and nonoscilla.tion of half~linear differential equations
145
If x'(t) > 0 for t 2 T, then from (3.7.21) we have x(~) ~ x(t) +

a l /a(t)x'(t)7r(t) for ~ 2 t 2 T, which shows that x(t) is bounded on
[T,oo). If x'(t) < 0 for t 2 T I , then x(t) is clearly bounded. Letting
~ --+ 00 in (3.7.21), we get 0 ~ x(t) + al/a(t)x'(t)n(t) for t 2 T. In
either case, we obtain n(t)al/a(t)x'(t)/x(t) 2 -1 for t 2 T from which
(3.7.18) is an immediate conseqnence.
The relation (3.7.19) trivially holds if x'(t) < 0 for t 2 T I , since in
this case the function defined in (3.7.17) itself is negative for t 2 T I . If
x'(t) > 0 for t 2 T, then there exist two positive constants kl and k2
such that x(t) 2 kl and a(t)1j;(x'(t)) ~ k2 for t 2 T, which imply
a(t) 1j;(X'(t)) < -k2

1j;(X(t)) - k?
Now, since n(t) --+ 0 as t --+
00,
t> T.
for
we conclude that
a
1j;(x'(t))
t~~ n (t)a(t) 1j;(x(t)) ~ O.
This proves (3.7.19) and completes the proof.

Theorem 3.7.4. Let ..\ 2
nonoscillatory if and only if
0:
+1
be a constant. Thcn cquation (3.1.1) is
(3.7.22)
and there exists a function y(t) E C([T, (0), JR), T 2 to such that
nQ(t)y(t) is bounded on [T, (0), nQ(t)y(t) 2 -1 for t 2 T (3.7.23)

and
nQ(t)y(t)
2l
oo
+0:
n),(s)q(s)ds +..\
loo a~I/Q(s)n)'~I(s)y(s)ds
loo a~I/Q(s)n),(s)ly(s)I(Q+I)/Qds
(3.7.24)
for t 2 T.
Proof. (The 'only if' part). Let x(t) be a solution of equation (3.1.1)
such that x(t) lOon [T, (0) for some T 2 to. Define w(t) =
a(t)1j;(x'(t))N(x(t)) on [T, (0). Then, we have
w'(t)
+ o:a~I/Q(t)lw(t)I(Q+I)/a + q(t)
on
[T, (0).
(3.7.25)
Multiplying (3.7.25) by n),(t) and integrating it over [t,~], ~ 2 t 2 T,
Chapter 3
146
we obtain
1(.
nA(~)w(~) -
nA(t)w(t) + A
a-1/0(s)nA-l(s)w(s)ds
+ l\r A(s)q(s)ds + 0: l(.a- l / Q(s)n A(s)lw(S)I(Ct:+l)/ods = 0, ~ ~ t ~ T.

Clearly, (3.7.24) follows from (3.7.26). Also nQ(t)y(t)
trivial from Lemma 3.7.3.
(The 'if' part). Let
y(t)
n-A(t)
+0:
[1
00
00
00
-1 for t
(3.7.26)
~ T is
a- l / O(s)nA-l(s)w(s)ds
a-1/Q(s)nA(s)lw(s)I(O+l)/QdS] .
Then by (3.7.24), w(t)

that
nQ(t)w(t)
nA(s)q(s)ds + A
y(t) for t
T. This and Lemma 3.7.2 imply
nQ(t)y(t)
~ nO-A(t) lOa-l/Q(s)nA-Q-l(s) [AnO(s)w(s) + o:lnO(s)w(s)l(o+l)/Q] ds

2: - ( -A- )0+1 nQ-A(t)
0:+1
00
a-l/o(s)nA-Q-l(s)ds
= - -1- ( -A- ) 0+1 nQ-A(t)nA-Q(t)

A-O:
0:+1
and 9 (nO(t)w(t)) ~ 9 (nQ(t)y(t))

defined in Lemma 3.7.2.
for t
1 ( -A- ) Q+l
A-O: 0:+1
=
~
- --
T, where the function 9 is
Since
y'(t)
for t
y'(t)
An-l (t)a-l/<t(t)w(t) - c(t) - h-l(t)a-l/Q(t)w(t)

_o:a- l / o (t)lw(t) I(Q+l)/o
T, we have
+ q(t) + o:a-l/O(t)ly(t)I(Q+l)/Q
An- l (t)a- l / o (t)y(t) _An-l(t)a- l / o (t)w(t) - o:a-l/a (t) Iw(t) I(o+l)/a
+ o:a-l/O(t)ly(t)l(o+l)/Q
a-l/o(t)n-o-l(t) {[o:lnO(t)y(t)l(o+l)/O
- [o:lnO(t)w(t)l(a+l)/o
+ AnO(t)y(t)]
+ AnO(t)w(t)]}
a-l/o(t)n-o-l(t) [9 (nO(t)w(t)) - 9 (nO(t)y(t))] < 0 for t ~ T.
147
Now, Theorem 3.7.1 guarantees that equation (3.1.1) is nonoscillatory.
3.7.2. Comparison Theorems

In this subsection, we shall consider (3.1.1) and the equation
(al(t)'lj;(y'(t)' +ql(t)'lj;(y(t
(3.7.27)
0,
where al(t) E CI([to,oo),JR+) and ql(t) E C([to,oo),JR).

We shall employ the nonoscillation characterization presented in Theorem 3.7.2 to prove the following comparison theorem.
Theorem 3.7.5. Let aCt) 2: al(t), limt-+=R(t) =
11= q(s)dsl
1=
ql(s)ds
00,
and
t 2: to
for
If (3.7.27) is nonoscillatory, then so is the equation (3.1.1).
Proof. Since equation (3.7.27) is nonoscillatory it follows from Theorem

3.7.1 that there exists wet) E CI([T,oo),JR) for some T 2: to such that
w'(t)
+ ql(t) + aa~l/a(t)lw(t)l(a+l)/a
~ 0
for
t 2: T.
Using the fact that aCt) 2: al (t) in the above inequality, we obtain
w'(t)
+ ql(t) + aa-I/O(t)lw(t)l(o+I)/o
~ 0
for
t 2: T.
Thus, by Theorem 3.7.1, we find that the differential equation
(a(t)'lj;(x'(t)))'
+ ql(t)'lj;(X(t =
(3.7.28)
is nonoscillatory. Now, it follows from Theorem 3.7.2 that there exists a

function yet) E C([T*,oo),JR) for some T* 2: T such that
ly(t)1
>
11=
>
11'' ' ' q(s)ds + a
1=
1=
ql(s)ds + a
a-I/O(s)ly(s)l(o+I)/O!dsl
a-I/O!(s)ly(s)l(o+I)/O!dsl
Once again, we can apply Theorem 3.7.2 to conclude that equation (3.1.1)
is nonoscillatory. This completes the proof.
Next, we shall consider equations (3.1.1) and (3.7.27) where q(t), ql(t)
EC([to,oo),JR+), and
1=
t
--ds
ales) .
(3.7.29)
Chapter 3
148
Using the nonoscillation characterization established in Theorem 3.7.4, we

shall prove the following comparison theorem.
Theorem 3.7.6.
Assume that a(t) 2: al(t) and (3.7.29) holds, and let

(3.7.30)
for all sufficiently large t, where A 2: 0'+1 is a constant. If (3.7.27) is

nonoscillatory, then so is the equation (3.1.1).
Proof. Since (3.7.27) is nonoscillatory it follows from Theorem 3.7.1 that
there exists a function w(t) E C 1 ([T, (0), lR) for some T 2: to such that
w'(t) + qdt) + O'a~1/0(t)lw(t)l(o+1)/o
0 for t 2: T. Hence, we have
w'(t) + q1(t) + c:w- 1/ O (t)lw(t)l(o+1)/o 0 for t 2: T. By Theorem 3.7.1,
equation (3.7.28) is nonoscillatory. Now, from Theorem 3.7.4 and (3.7.30),
we conclude that (3.1.1) is nonoscillatory. This completes the proof.
:s:
:s:
3.7.3. Sufficient Conditions for N onoscillation

of Equation (3.1.1)
Another version of Theorem 3.7.1 is the following result.
exist T 2: to and a function w(t) E C 1 ([T, (0), lR) such that
q(t)
+ O'a(t)lw(t)l(o+l)/a - (a(t)w(t))' :s:
on
[T, (0).
(3.7.31)
The proof of Theorem 3.7.7 is similar to that of Theorem 3.7.1, except

that now we let w(t) = '!j;(x'(t))N;(x(t)), t 2: T. The details are left to
the reader.
As applications of Theorem 3.7.7, now we present the following nonoscillation criteria for the equation (3.1.1). In what follows, we shall let by
,= (1/0') (0'/(0' + l)t+
1.
Theorem 3.7.8. Let h(t) E C 1 ([to,00),lR+) and cf>(t) E C 1 ([to,00),lR)

satisfy
(3.7.32)
and
:s: -q(t),
2: to
(3.7.33)
lim sup hO(t)Icf>(t)1 < "
(3.7.34)
cf>'(t)
If
t-+oo
149
Proof. Inequality (3.7.34) implies that there are two numbers T

and kE(O,-y) suchthat 1(t)l:Skh-a(t) for t~T. Let
w(t)
where A = ((a
+ l)/a)O;.
Aa(t)
_ A,.'()
'+'
- Ak)
+ 1hO;(t)
,
(_a )0;+1
1
< 1,
a+1
a+ 1
A(t)
to
Then, we have
= (~)O;.!.
(a(t)w(t))'
a(l - Ak) h'(t)

A
hO+l(t) > q(t)
(3.7.35)
a(l - Ak) a- 1/0;(t)

A
hO+1(t)
and
q(t)
+ aa(t)lw(t)I(O;+l)/O; - (a(t}w(t))'
:S q(t)
1-Ak) 1(0;+1)/0;
a(l-Ak) a- 1 / o (t)
-q(t)A
h a+1(t)
+ aa(t) 1Aa(t) A(S) + h<>(t)
= aa-1/O(t)h-(o+1)(t) [I(t)ha(t) +
:Saa-1/<>(t)h-(<>+l)(t)
ra+1)/a _
~ Ak]
[Ik+ I~Akl(a+1)/<> _1-/k]
= aa- 1/<>(t)h-(o+1)(t) (
=
1 -A Ak
A(o+l)/o
.!.A + k)
aa- 1/<>(t)h-(<>+1)(t)(k - -y) :S O.

Theorem 3.7.9.
be replaced by
Conditions (3.7.32) and (3.7.33) of Theorem 3.7.8 can
h'(t) :S - a- 1/0;(t)
for
t ~ to
(3.7.36)
and
'(t)
q(t)
for
t ~ to,
(3.7.37)
respectively.
The proof of Theorem 3.7.9 is the same as that of Theorem 3.7.8 except
that now, we let
1
w(t) = Aa(t)
1- Ak)
A(t) + h<>(t) .
150
Chapter 3
Theorem 3.7.10. Let the functions h(t) and (t) be as in Theorem 3.7.8
such that conditions (3.7.32) and (3.7.33) hold. If there exists a number
k > 0 such that
_ko)(o+l) -- k
:s
hOo(t)I(t)1
:s
kOo/(o+l) - k < ,,(,
(3.7.38)
then equation (3.l.1) is nonoscillatory.

Proof.
Let
w(t)
1 ( (t)
a(t)
k )
+ hOo(t)
Then, we have
(a(t)w(t))'
,h'(t)
- (t) + ak h a+1(t) > q(t)
a- 1/"'(t)
+ ak h a+1(t)
and
q(t)
+ aa(t)lw(t)I("'+l)/c>
1
< q(t) + aa(t) 1a(t)
- (a(t)w(t))'
(t)
+ h"(t)
aa- 1 /O:(t)h-(o:+1)(t) [1(t)hO:(t)
< aa- 1/O:(t)h-(o:+1)(t)(k - k)
) 1("+1)/00
a- 1 /a(t)
- q(t) - ak hO:+ 1(t)
+ kl(a+1)/a
- k]
o.
Now, it follows from Theorem 3.7.7 that (3.l.1) is nonoscillatory.
Theorem 3.7.11. Conditions (3.7.32) and (3.7.33) of Theorem 3.7.10 can

be replaced by (3.7.36) and (3.7.37), respectively.
The proof of Theorem 3.7.11 is the same as that of Theorem 3.7.10

except that now, we let
w(t) = att) ((t)
+ hQk(t))
Example 3.7.1. Consider the half-linear equation
(3.7.39)
where f3 > a
and () > 0 are constants. Let h(t)

()s(3-Oo- l ds. Then for t 2': to > 0,
(t)
h(t)
~t(o:-(3)/""
f3-a
(t)
= _()_t(3-a
f3-a
= ftoo s-f3/ Qds and
and h'" (t)(t)
= -()- (~) '"

f3-a
f3-a
151
If () ~ ((f3 - 0.)/(0. + 1))'~+1, then h"'(t)<f>(t) ~,. Now, it follows from

Theorem 3.7.11 that (3.7.39) is nonoscillatory.
Theorem 3.7.12. Let h(t) E C 1([to,00),JR+) be such that condition
(3.7.32) hold. If there exists (t) E C 1 ([to, (0), JR) such that
lim <f>(t)
exists and
t-'too
'(t) ~ - h"'(t)q(t),
(3.7.40)

Proof.
Since limt-'too (t) exists, there are two numbers T 2: to
and m such that 0 < m + (t) ~ 1 for t 2: T. Let w(t) = -(m +
<f>(t))/(a(t)h"'(t)), t 2: T. Then for t 2: T, we have
(a(t)w(t))'
h'(t)
'(t)
a- 1 / Q (t)
o.(m+(t)) hQ+1(t) - hQ(t) > q(t) +o.(m+(t)) hc>+l(t) ,
which implies
+ o.a(t)lw(t)I(Q+1)/c> - (a(t)w(t))'
m + (t) 1(<>+1)/C>
a- 1 /<>(t)
~ q(t) + o.a(t) 1a(t)hc>(t)
- q(t) - o.(m + (t)) hc>+l(t)
< o.a- 1/ Q(t)h-(c>+1) (t)[(m + <f>(t)) - (m + (t))] = O.
q(t)
Theorem 3.7.13. Condition (3.7.32) in Theorem 3.7.12 can be replaced

by (3.7.36).
Proof.
Since limt-'too ( t) exists, there are two numbers T 2: to
and m such that 0 < m - <f>(t) ~ 1 for t 2: T. Let w(t) = (m(t))/(a(t)hC> (t)). The rest of the proof is similar to that of Theorem 3.7.12
and hence omitted.
Theorem 3.7.14. Let h(t) E C 1 ([to,00),JR+) be such that q(t)hQ+1(t) ~

1. If either
for all sufficiently large t,
lim (h'(t) - a- 1/c>(t))
= exists and
or
1
a
h'(t) - a- 1 / (t) 2: -
t-'too
2: 1/0.,
(3.7.41)
(3.7.42)

Proof. It follows from (3.7.41) or (3.7.42) that there is a number T 2: to
such that h'(t)-a- 1/ Q(t) 2: I/o. for all t 2: T. Let w(t) = -l/(a(t)hQ(t)).
152
Chapter 3
Then, we have
q(t)
+ aa(t)lw(t) 1(<>+1)/<> - (a(t)w(t))'

1
::; h(}+l(t)
+a
ah-(<>+l)(t)
a- 1/<>(t)
h'(t)
h<>+1(t) - a h<>+l(t)
(~ + a- 1/<>(t) -
h'(t))
::; 0
for
Hence, by Theorem 3.7.7, equation (3.1.1) is nonoscillatory.

Theorem 3.7.15.
(3.7.33) hold. If
00
t 2: T.
Let </>(t) E C1([to,oo),lRo) be such that condition
a- 1/<>(s)</>(cx+1)/CX(s)ds <
--
) (<>+1)/<>
a+1
</>(t) ,

Proof. Let
w(t)
__1_
a(t)
[</>(t)
+ a(a + 1)(0+1)/<>
roo a- 1/
it
(s)</>(a+1)/<> (S)dS] .
Then, we have
which implies
q(t)
+ aa(t)lw(t)I(<>+1)/<> - (a(t)w(t))'
~+1
::; q(t)+aa(t)
a~t) (</>(t) +a( a+ 1)(<>+1)/<>l":-l/"(s )</>("+1)/" (s )ds)1 <>
- q(t) - a(a+1)(,,+1)/"a- 1/"(t)</>(<>+1)/"(t)

= aa- 1/<>(t)
rooa- 1/ cx (s)</>(,,+1)/cx(s)ds 1(,,+1)/<>
[I </>(t) + a(a + 1)(<>+1)/" it
-(a + 1) (cx+1)/cx</>(,,+1)/cx (t)]

::; aa-1/"(t)
[I </>(t) + a(a + 1) (cx+1)/<> (a + l)-(,,+1)/cx</>(t) 1(a+1)/cx
-(a + 1)(<>+1)/"</>(<>+1)/,, (t)]
= o.
153
Oscillation and nonoscillation of haJE-linear differential equations
Theorem 3.7.16. Let r/>(t) E C 1([to, 00), lR o) satisfy condition (3.7.33),

and let
r/>1 (t) = 1a-1/0(s}:P(0+1)/0(s) exp [a(a + 1)1/01 8 a-l/0(~)r/>1/0(~)d~] ds.

If r/>l(t) ::::: (a + l)-l/r/>(t), t ~ to then equation (3.1.1) is nonoscillatory.
Proof. Let w(t) = - [r/>(t)
+ a(a + l?/r/>l(t)] /a(t).
Then, we have
(a(t)w(t))' = -r/>'(t) - a(a + l)l/o.r/>~(t)
~ q(t) + a(a + l)l/o.a- l /o.(t) [a(a+1?/0.r/>1/0.(t)r/>1(t) + r/>(a+1)/0.(t)] ,

which implies
q(t) + aa(t)lw(t)l(o+1)/a - (a(t)w(t))'
< q(t) + aa- 1/0.(t) [r/>(t) + a(a + 1?/0.r/>1(t)] (0.+1)/0. - q(t)

- a(a
+ l)l/o. a-l/o(t)
aa-l/o.(t) [r/>(t)
x
{[r/>(t)
+ r/>(o.+l)/o.(t)]
+ a(a + l)l/Or/>l(t)]
+ a(a + 1)1/r/>1(t)] I/o
::::: aa-l/o(t) [r/>(t)

x
[a(a + l)l/o.r/>l/o.(t)r/>l(t)
- (a
+ 1)1/0r/>1/0(t)}
+ a(a + l)l/r/>l(t)]
{[r/>(t) + a(a + l)l/o(a + l)-l/Or/>(t)] I/o - (a + 1)1/0r/>1/0(t)}
aa-l/a(t) [r/>(t) + a(a + l)l/ar/>l(t)] [(a + l)l/o.r/>l/o.(t)
-(a + l)l/o.r/>l/o.(t)] =
o.
Theorem 3.7.17. Let r/>(t) E C 1([to,00),lR+) be such that r/>'(t)

a- 1/0.(t). If there exists a T 2:: to such that
11
00
r/>o.(s)q(s)dsl < 00 on

Proof. Let
[T,oo),
(3.7.43)
154
Chapter 3
Then, we have
In view of (3.7.43) there exists a T1 ::;> T such that
o<
1 + 21
00
cjJ<>(s)q(s)ds :::; 2 for all
t::;> T1 .
Thus, we find
2-(0+1)/0cjJ-(0+1)(t) [1 + 21 00 cjJO(S)q(S)dS]
ly(t)l(a+1)/0
<
~cjJ-(Q+1)(t)
[1+21
00
cjJ<>(s)q(S)dS].
Consequently, (3.7.44) implies that (3.7.1) holds on [T1' 00). Now, Theorem
3.7.1 guarantees that equation (3.1.1) is nonoscillatory.
3.7.4. Further Results on Nonoscillation

of Equation (3.1.1)
Here, we shall consider the following half-linear differential equation
where R(t) E C 1([to,00),JR+) satisfies R'(t) = a- 1 / 0 (t) and limHoo R(t)

= 00.
Clearly, (3.7.45) has a solution x(t) = Rm(t) on [to, 00) if m

satisfies the indicial equation f(m) = o:m(m - 1)lmI 0 - 1 + ), = o. If
), = (0:/(0: + 1))0+1, then (3.7.45) is nonoscillatory because it has a
nonoscillatory solution y(t) = Ro/(<>+l)(t). Equation (3.7.45) is oscillatory if ), > (0:/(0: + 1))0+1 because then f(m) = 0 has no real roots.
Further, (3.7.45) is nonoscillatory if ),:::; (0:/(0: + 1))<>+1 because then
f(m) = 0 has a real root.
As an application of Theorem 3.7.5, we have the following result.
Theorem 3.7.18. Let !too q(s)ds < 00 and R(t) E C 1 ([to, 00), JR+)
satisfy R'(t) = a- 1 /<>(t) and limHoo R(t) = 00. Then,
(i)
equation (3.1.1) is nonoscillatory if
RQ(t)
00
q(s)ds:::; _1 ( _0:_ ),,+1

0: 0:+1
for all sufficiently large
t,
155
equation (3.1.1) is oscillatory if there exists a number ,X* such that
(ii)
RO(t)
00
q(s)ds > ,X* >
have
00
!..O! (_0!_)0+1
0!+1
t.
1 ( O! )0+1
-R-O(t).
O! 0!+1
q1(s)ds = -

(ii) By assumption there exists a T 2': to such that
Since equation (3.7.45) with ,X = O!'x* is oscillatory, it follows from Theorem

3.7.5 that (3.1.1) is oscillatory.
Next, we present another useful version of Theorem 3.7.2.

exists a function c( t) E C ([T, 00), JR) for some T 2': to such that
Ic(t)+Q(t)1 2': iQ(t) + O!
00
a- 1/ O(s)lc(s) + Q(S)I(0+1)/Odsi
for t 2': T.
(3.7.46)
Corollary 3.7.1.
If Q(t) 2': 0 and there exists a function
C ([T, 00), JR +) for some T 2': to such that
c(t) 2': O!
00
a- 1/ O(s)lc(s) + Q(s)I(0+1)/ods
for
2': T,
c(t)
(3.7.47)

Let b(t) E C([to, 00), JR) and c(t) = b+(t) - Q(t), where b+(t) =
maxi b( t), O}. Then by Theorem 3.7.19, we have the following corollary.
Corollary 3.7.2. If there exists a function b(t) E C([T, oo),JR) for some
T 2': to such that

Next, we let b(t) = (O!+I)IQ(t)1 for t2':T2':to in Corollary 3.7.2, to
obtain the following result.
Chapter 3
156
Corollary 3.7.3. If
00
a-l/<>(s)IQ(s)I{<>+l)/<>ds <
a+1
) (a+l)/<>
IQ(t)l,

To obtain the next result, we let c(t) = k<>/{<>+l)R<>(t) - Q(t) where
k > 0 is a constant. Then, we have Ic(t) + Q(t)1 = k<>/(<>+l) R-<>(t) and
Corollary 3.7.4. If there exists k > 0 such that
a_)
<>+1 ,
_k<>/{<>+l) - k :::; R<>(t)Q(t) :::; k<>/{<>+l) - k :::; _1 ( __
a a+1
Now, taking yet) = Ql(t)+b(t) in (3.7.13) for some bet) E C([T, oo),lR)
where T ~ to, we have the following nonoscillatory characterization result.
Theorem 3.7.20. Let I,ds, t], Q(t) and Ql(t) be as in Theorem 3.7.3.
Then equation (3.1.1) is nonoscillatory if and only ifthere exists a function
bet) E C([T, 00), lRo) for some T ~ to such that
bet)
00
{IQ(s)+a[Ql (s)+b(s)]I{<>+l)/<>-(a+ l)IQ(sW/<>[Ql (s) +b(s)]
x sgn Q(s) -IQ(s)I{<>+l)/<>} a-l/<>(s)/L[S, t)ds.

(3.7.49)
Let bet) = (,X - l)Ql(t), ,X > 1 in (3.7.49), to obtain the following
corollary.
Corollary 3.7.5. Let p,[s, t], Q(t) and Ql (t) be as in Theorem 3.7.3.
If there exists a constant ,X > 1 such that
(,X - l)Ql(t)
00
{IQ(s)
+ a'xQl (s) I{<>+l)/<> -
(a
+ l),XIQ(sW/<>
x Ql(S) sgn Q(s) -IQ(s)I{<>+l)/<>} a- l /<> (s)p,[s, t)ds,

If we let ,X = (a + l)/a in Corollary 3.7.5, then an easy computation
gives the following result.
157
Corollary 3.7.6. If Q( t) ?: 0 for t?: to and

Corollary 3.7.7. If Q(t)?: 0 for t?: to and
1=
Cl!~l)
Ql/OI(s)a- l / OI (s)ds :::;
[1- (0:1)01] Q(t),
(3.7.50)

Proof. Let
Multiplying both sides of (3.7.50) by
ft"o
Ql/OI (t)a- l / OI (t)

(?: 0 by assumption),
Q(OI+l)/OI(s)a- l / OI (s)ds
and integrating it over [s, t], we obtain
Thus, we have
<
1=
Q(OI+!)/<>(s)a- l / OI (S)J.ll[S,tjds
1
r= Q(<>+l)j<>(s)ds <
l-'Y(o+l)it
o
~ 1 (( 0:
1)01 - 1) Q(t).
Now, it follows from Corollary 3.7.6 that (3.1.1) is nonoscillatory.
The following result improves Theorem 3.7.20.

Theorem 3.7.21. Let <f>(t) E Cl([to, 00), lR) be such that <f>'(t) :::; -q(t)
for t?: to, and let
Chapter 3
158
where ). is a constant. If one of the following conditions holds
(i)
).>1 and 4>(t) ~
(ii)
)'~1
a).
).0 _
1 ;3(t),
and 4>(t) :::: -a).;3(t),
(iii) ).<1 and
a).
IAI
A;3(t) :::: 4>(t) :::: -aA;3(t),
-1 +
0 sgn

Proof. Since one of the conditions (i), (ii) and (iii) holds,
14>(t) +aA;3(t)IC a + I)/a - A{14>(t)I Co+I)/a + aAI4>(t)I I /;3(t) sgn 4>(t)} :::: O.
Let wet)
= -[4>(t) + aA;3(t)]/a(t). Then, we have
(a(t)w(t))' = - 4>'(t) - a)..;3'(t)
~ q(t) + a)..a-I/o(t) {14>(t)I Ca+I)/a + a)..I4>(t)I I /;3(t) sgn 4>(t)} ,

which implies
q(t)
+ aa(t)lw(t)I,t' - (a(t)w(t)'
-A [14>(t)1 0;;'
aa-I/a(t) {14>(t)
+ aAI4>(tW/;3(t)
+ aA;3(t) I0;;'
sgn 4>(t)]} < O.
Similar to Corollary 3.7.7, we have the following result.

Corollary 3.7.8. Let 4> E CI([to,oo),lR) be such that 4>'(t):::: -q(t), t ~
to. If there exists a constant ).. > 1 such that

Remark 3.7.1. Let Q(t) = 4>(t) ~ 0 for t ~ to. If ).. = (a+l)/a, then
Theorem 3.7.21 and Corollary 3.7.8 reduce to Corollaries 3.7.6 and 3.7.7
respectively.
Now, we shall consider the following half-linear differential equation
(a(t)lx'(t)IO-Ix'(t))' + Aa- I / O(t)n- Cc<+1) (t)lx(t)IO-Ix(t) = 0,

where A is a constant, net) = ftOO a-I/cs)ds and n(to) <
00.
t ~ to
(3.7.51)
159
Clearly, (3.7.51) has a solution x(t) = 7r- m (t) for t:::: to if m

satisfies the indicial equation g(m) = o:m(m - l)JmJ<>-l + A. If A =
(0./(0: + 1))<>+1, then (3.7.51) is nonoscillatory because it has a nonoscillatory solution x(t) = 7r o /(o+1) (t). Equation (3.7.51) is oscillatory if
A> (0./(0: + 1))<>+1 because then g(m) = 0 has no real roots. Further,
(3.7.51) is nonoscillatory if A:::; (0./(0. + 1))<>+1 because then g(m) = 0
has a real root.
By Theorem 3.7.6 and a technique similar to the one used in Theorem
3.7.16, we have the following result.
Theorem 3.7.22. Let q(t) E C([to,oo),ffi+) and 7r(to)
(i)
< 00. Then,
1)
_(
7rt
(ii)
00
7r0+1(s)q(s)ds:::;
()<>+1
_0._
0:+1
t,
equation (3.1.1) is oscillatory if there exists a number A* such that
-1()
7rt
00
7rct+1(s)q(s)ds :::: A* >
_0._ )0+1 for all sufficiently large t.

0:+1
The following corollary is an immediate cO:J.sequence of Theorem 3.7.22.

Corollary 3.7.9. Consider the following half-linear differential equation
(3.7.52)
where ,> 0: and A > 0 are constants. Then,
+ 1))<>+1 ,
(i)
equation (3.7.52) is nonoscillatory if A:::; (b - 0.)/(0.
(ii)
equation (3.7.52) is oscillatory if A> (b - 0.)/(0: + 1))"+1 .
Proof. From equation (3.7.52) it follows that
,-a
_o._t(<>--Y)/O
and
(_a
,-a
)0+1
Now, the result follows from Theorem 3.7.22.
In an analogous way we can extend Corollary 3.7.9, to obtain

Corollary 3.7.10. Consider the following half-linear differential equation
(fYJx'(t)J
O -
x'(t))' + q(t)Jx(t)Jo-1 X(t)
0,
t:::: to
(3.7.53)
160
Chapter 3
where
(i)
'I > a
is a constant. Then,

tb-o:)/o:
00
< (
s-b-o:)(o:+l)/u q (s)ds
a ) ('I~
~ a) 0:
for all sufficiently large t,

(ii) equation (3.7.53) is oscillatory if there exists a number A* such that
tb-o:)/o:
1=
> A* > (
s-b-o:)(e>+l)/D q (s)ds
a )
0'+1
('I ~ 0')0:
0'+1
for all sufficiently large t.

Next, by using an approach different than those given above, we shall
present nonoscillation criteria for the more general equations
+ F(t, x(t))
(3.7.54)
+ q(t)f(x(t))
0,
(3.7.55)
(lx'(t)l a - 1x'(t))'
and
(lx'(tW-1X'(t))'
where
a > 0 is a constant,
(i)
(ii) FE C([to, 00) x IR, IR), sgn F(t, x) = sgn x for each t E [to, 00),
(iii) f E C(IR,IR), sgn f(x) = sgn x, and
(iv) q(t) E C([to,oo),IR+).
Theorem 3.7.23. Suppose F(t,x) is nondecreasing in x for each fixed
t and has the partial derivative with respect to t such that
of
sgn Tt(t, x)
~ sgn
for
t 2': to
(3.7.56)
Furthermore, suppose that
j= IF
and
(s, cs) Ids
<
001
00
for all nonzero constant c
1m sup
\--+0
F(S,AS)d
\
AO:
<
00,
(3.7.57)
(3.7.58)
Proof.
{td~l
Suppose equation (3.7.54) has an oscillatory solution x(t). Let

be an increasing sequence of zeros of x(t) such that X'(tk) > 0
161
for k = 1,2, .. '. Let Tk be the first zero of x'(t) which is located to the
right of tk' Since x'(t) > 0 on (t k , Tk) and x'(t) is decreasing there,
we have
x(t) = X(tk)
+ rt x'(s)ds :::; X'(tk)(t - tk) :::; ty'(tk)
for
Jtk
t E
(tk' Tk)'
(3.7.59)
Define
a
V(t) = --lx'(t)I"+l
a+1
x (t)
F(t,u)du
for
t 2:: tl.
(3.7.60)
Clearly, V(t) 2:: 0 for t 2:: t l , and
V'(t)
(lx'(t)I",-IX'(t))' x'(t)
rCt)
+ F(t, x(t))x'(t) + J o
rct) of
Jo
7it(t,u)du:::; 0 for
of
7it(t, u)du
t 2:: tl'
This implies that V(t) is bounded for t 2:: tl' Thus, in view of (3.7.60),
x'(t) remains bounded as t ---+ 00. It follows that there exists a constant
M > 0 such that
(3.7.61)
Ix'(t)1 :::; M for t 2:: tl
Now, we integrate (3.7.54) from tk to Tk, and use (3.7.59) and (3.7.61),
to obtain
r k F(s, x(s))ds < r k F(s, x'(tk))ds

Jtk
Jtk
<
roo
F(s, Ms)ds,
rkF(s,Ms)ds:::;
Jt k
Jtk
which implies that X'(tk) ---+ 0 as k ---+
r k F(s, SX'(tk))
(X'(tk))" ds
1 :::; Jtk
and
00,
for
= 1,2,.
(3.7.62)
Let T > tl be fixed arbitrarily and choose tk so that tk 2:: T. Then,

it follows from (3.7.62) that
1
<
-
roo
r k F(s, SX'(tk)) ds <

F(s, SX'(tk)) ds
JT
(X'(tk))"
- JT
(X'(tk))"
and hence by Fatou's lemma as T ---+
00,
Chapter 3
162
which is a contradiction. Hence, equation (3.7.54) cannot have an oscillatory solution.
Suppose f(x) is nondecreasing, q(t) E CI([to, 00),
Theorem 3.7.24.
lR+) and satisfies
00
[q'(s)]+
ds <
q()
s
00,
[q'(t)]+ = max{q'(t) , O}.
(3.7.63)
for all nonzero constant c
(3.7.64)
Furthermore, suppose that
Joo q(s)lf(cs)lds
and
<
00
JOO q(s)li~~p I~~~~)I ds
<
(3.7.65)
00.
Then equation (3.7.55) is nonoscillatory.

Proof. Let x(t) be an oscillatory solution of (3.7.55) on [to, 00). We
will show that x'(t) remains bounded as t -+ 00. Define the function
V (t) as follows
a
Vet) = --lx'(t)ICt+1
a
+1
+ q(t)
(t)
f(u)du,
t 2: to
Differentiation of Vet) gives
V'(t)
r(t)
q'(t) io
r(t)
f(u)du::; ([q'(t)]+) io
f(u)du,
which implies that V'(t) satisfies the first order differential inequality
V'(t) ::; [q'(t)]+ Vet)

q(t)
for
t 2: to.
Hence, we have
Vet) ::; Veto) exp
(it [q'(s)]+
dS),
s)
to
q(
t 2: to.
Thus, Vet) is bounded. The definition of Vet) now shows that x'(t)
remains bounded as claimed. Proceeding as in Theorem 3.7.23, we get the
required contradiction.

( l
x'(t)ICt-IX'(t)' + Ix(t)I.6- I X (t) = 0,

t"l(t 2 + x 2(t8
t 2: 1
(3.7.66)

where
0:,
{3, ,
163
and 8 are positive constants. The function
is nondecreasing in x if {3 2': 28 and satisfies (3.7.56). The condition

(3.7.57) is satisfied if {3~28+1 < ,,/, whereas (3.7.58) is satisfied if 0::::; {3
and (3 ~ 28 + 1 < 8. Consequently, Theorem 3.7.23 implies that all solutions
of (3.7.66) are nonoscillatory if {3 2': max{ 0:, 28} and "/ > {3 ~ 28 + 1.
Similarly, it can be shown that all solutions of the equation
(lx'(t)I O- 1x'(t))' + e- tf3 sinh (lx(t)IO-1X(t)) = 0,
(3
> 0: > 0 (3.7.67)
are oscillatory.

Here, we shall consider (3.1.1) with a(t)
1, i.e., the equation
(lx'(t)I,,-l X'(t))' + q(t)lx(t)I"-lX(t) = 0,

where 0: > 0 and q(t)
the linear equation
C([to, 00), JR.). When

x"(t)
+ q(t)x(t)
0:
(3.8.1)
1, (3.8.1) reduces to
o.
(3.8.2)
When 0: of- 1, although equation (3.8.1) is nonlinear it has the property

that any constant multiple of a solution is also a solution. The term 'halflinear' is used to denote this property.
In the oscillatory theory of (3.8.1) there are several results which are
similar to those known for the equation (3.8.2). We state some of these in
the following theorems.
Theorem 3.8.1. Equation (3.8.1) is oscillatory if
liminf t"+1q(t) >

t->oo
0:
(_
0:
+1
and is nonoscillatory if
0:
limsup t"+lq(t) < ( _
t->oo
0:
+1
),,+1
),,+1
Another form of Theorem 3.8.1 is the following result.
164
Chapter 3
Theorem 3.8.1'. Equation (3.8.1) is oscillatory if there exists a constant

to > 0 such that
ex: 1)
t<>+lq(t) :::::
<>+1
+ to
for all sufficiently large t
and is nonoscillatory if
Theorem 3.8.2. Suppose q(t) > 0 for t::::: to ::::: O. Then equation
(3.8.1) is oscillatory if either one of the following conditions holds
(i)
(ii)
q(t) (j. .c[to, (0), or

q(t) E .c[to, (0) and
liminf t<>
t ..... oo
00
q(s)ds >
(0:
oP
+ 1)<>+1
and equation (3.8.1) is nonoscillatory if q(t) E .c[to, (0) and

lim sup to/
t ..... oo
00
q( s )ds <
0:0/
(0:+1)<>+1'
Theorem 3.8.3 (Generalized Sine Function). Let S(t) denote the

solution of (3.8.1) with q(t) = 0: determined by the initial conditions
x(O) = 0, x'(O) = 1. Then, S(t) exists on IR and for t E IR has the
properties
S(t
where 11"0/
+ 11"0/) = S(t)
and
IS(tW+1
+ IS'(tW+1 =
1,
211"
.
(0: + 1) sin(1I"/(0: + 1)
Theorem 3.8.4 (Generalized Priifer Transform). Let x(t) be a solution

of (3.8.1) on [to, (0). Define the functions p(t) and (t) by
x(t)
p(t)S((t,
x'(t)
Then, p(t) > 0 and (t) E C 1 ([to, (0), IR)

differential equations
p(t)S'((t).
and satisfy the following
p'(t)
p(t)
and
'(t)
IS'((t))10/+ 1
+ ~q(t)IS((t)W+1.
0:
(3.8.3)
165
Now, we shall state a result for the equation

(lx'(t)la~lX'(t))' + F(t, x(t)) = 0,
(3.8.4)
where a> 0 is a constant, FE C([to,oo) x IR,IR) and sgn F(t,x) =

sgn x for t E [to, (0), to 2': O.
Theorem 3.8.5.
that
mf
x#o
Suppose there exists q(t)
(i)
IF(t,x)
x I 1x
C>~
(ii) Suppose there exists q(t)

x#o
F(t,x)
~l
x a X
II
C([to, (0), IR) such
()
2': q t for all sufficiently large t.
If (3.8.1) is oscillatory with q(t)

oscillatory.
sup
:<::;
q(t)
(3.8.5)
as in (3.8.5), then equation (3.8.4) is
C([to, (0), IR) such that

(3.8.6)
If (3.8.1) is nonoscillatory with q(t) as in (3.8.6), then equation (3.8.4) is

nonoscillatory.
(3.8.7)
for t 2': 1, where a, b, (3, A and J-t are positive constants. Since
and
sup
b+ti3 lxI A
< -
x#o a+tl-'lxlA -
a'
(3
< J-t
for all large t, it follows from Theorems 3.8.1' and 3.8.5 that
(i)
equation (3.8.7) is oscillatory if (3 > J-t and b/a> (a/(a
+ 1)t+1 ,
(i) equation (3.8.7) is nonoscillatory if (3 < J-t and b/a:<::; (a/(a
+ l))a+1 .
Finally, we shall prove two results for the more general equation, namely,
(a(t)lx'(t)la-1x'(t))' + F(t, x(t))
0,
where a and F are as in (3.8.4) and a(t) E C([to, (0), IR+).
(3.8.8)
166
Chapter 3
Theorem 3.8.6.
00.
Suppose condition (3.8.5) holds and
If
lim
t-+oo
it
to
q(s)ds =
JO
a- 1/ n (s)ds =
(3.8.9)
00,

Let x(t) be a nonoscillatory solution of equation (3.8.8), say,
x(t) > 0 for t:::: to :::: O. Define
Proof.
w(t)
a(t)
Ix'(t)IO- 1 x'(t)
()
xQ t
for
(3.8.10)
t:::: to.
Then, we have
O.
t:::: to.
(3.8.11)
Thus, from (3.8.11) it follows that w'(t) +q(t) ::; 0 for t:::: to. Integrating
this inequality from to to t, we obtain 1L(t)-w(to)+ft:q(s)ds::;O, t::::
to. Hence, w(t) ---c> -00 as t ---c> 00, and therefore x'(t) < 0 for t:::: t1
for some sufficiently large t1 :::: to Choose t2 :::: t1 so that
q( s )ds > 0
for t > t 2 . Integrating (3.8.8) from t2 to t, we get
-
fL
(3.8.12)
Now, from the relation
fL
we find that
q(s)(xO(s))ds :::: 0 for t:::: t2' Thus, from (3.8.12) we
can conclude tha:-t a(t)lx'(t)10-1x'(t)::; a(t2)lx'(t2)10- 1x'(t2), t:::: t2 and
hence
(3.8.13)
An integration of (3.8.13) shows that x(t) ---c> -00 as t ---c> 00, which contradicts the fact that x(t) > 0 eventually. This completes the proof.
Theorem 3.8.7.
Suppose condition (3.8.5) holds, DC:::: 1 and

lim
t-+oo
i (i8 a(u)du)-1/0 ds =
t
to
to
00.
(3.8.14)

If
t-lim
'>oo -litis
to q(u)duds
t
to
00,
to ~ 0
167
(3.8.15)

Proof.

x(t) > 0 for t ~ to ~ O. Define the function w(t) as in Theorem 3.8.6
to obtain (3.8.11). Using condition (3.8.5) and integrating the resulting
inequality twice from to to t ~ to, we get
lit w(s)ds+-t ltls

q(u)duds
to a- 1/a(u)lw(u)l(a+1)/aduds+-litiS
t to to
a
to
to
(3.8.16)
where Co > 0 is a constant independent of t.

Let V(t)
V(t) <
It: Iw(s)lds.
(L
a(s)ds
:<::: co,
Then by Hoder's inequality, we have
) 1/(a+1) (
10 a- 1/a(s)lw(s)ICa+1)/a ds)
t
a/(a+1)
or
(
10 a(s)ds )-l/a v(a+1)/a(t)

t
:<:::
lto
a- 1/a(s)lw(s)l(a+1)/ads for t
From (3.8.15) and (3.8.16) it follows that there exists a t1
litto Iw(s)lds + -at ltls

to to a- 1/a(u)lw(u)l(a+1)/aduds
(3.8.17)
to such that
< 0 for t
--
~ to.
~ t1.
(3.8.18)
Using (3.8.17) in (3.8.18), we get
V(t)
it
to
---+-
r/
v(a+1)/a(s)
(It: a(u)du
ds < 0,
v(a+1)/a(t)
Let
W(t)
Itot a(u)du )l/a
Chapter 3
168
Then, we have
W'(t)
;::: 0:(<>+1)/<>
W(<>+l)/<>(t)
(lot
a(s)ds) -1/<>
t;::: t2'
for
(3.8.19)
Integrating (3.8.19) from t1 to t, we obtain for t;::: tb

0:<>+1
-
i (1S
t
t,
to
W (t)
Wet,)
a(u)du
-1/<>
ds:S
it W(<>+l)~<>(s)
W'()
t,
ds
dv
[
]
= 0: W- 1/<>(t1) - W- 1/<>(t)
v(<>+l)/<>
:S o:W- 1/<>(td
it (1
lim
t-+oo
t,
to
a(U)du)-l/<> ds :S 0:-aw- 1/ a (t 1)
<
00,

3.9.1. Oscillation Theorems for Equation (3.1.1)
Here, we shall employ the integral averaging techniques to obtain oscillation criteria for equation (3.1.1). In what follows for convenience, we shall
let by 0:0 = (0: + 1)<>+1. Our results of this subsection extend Theorems
2.5.6 - 2.5.13.
Theorem 3.9.1. Suppose there exists a continuous function H: D
{(t, s) : t ;::: s ;::: to} ---+ IR such that
(t, s) ED, t > s
(3.9.1)
- 8H(t, s)/8s is nonnegative and continuous on D.
(3.9.2)
H(t, t) = 0
for
t;::: to,
H(t, s) > 0
for
and
h(t, s)
If
.
hmsup H(
t-+oo
t, to
1t [
to
H(t, s)q(s) -
a(s)ha+l(t, S)]
Ha()
ds
0:0
t, s
00,
(3.9.3)
Oscillation and nonoscill8,tion of half-linear differential equations
169
Let x(t) be a nOl1oscillatory solution of (3.1.1). Assume that

x(t) =I- 0 for t ~ to. We define w(t) = a(t)lx'(t)la- 1x'(t)/(lx(t)la- 1x(t
Proof.
for t
to. Then for every t
w'(t) =
and consequently,
H(t, s)w'(s)ds
to
to, we have
q(t) - a
Iw(t)j(<>+l)/<>
a1/<>(s)
it
to
Since
H(t, s)w'(s)ds
- w(to)H(t, to) -
equation (3.9.4) gives
H(t, s)q(s)ds
S;
H(t, to)w(to)
to
-a
to
. to
H(t, s)
Iw(s)I(<>+1)/<>
1/<>() ds.
a
s
(3.9.4)
f)
W(S)-f) H(t, s)ds,
(3.9.5)
to
to
l.t
H(t, s)q(s)ds - a
it
h(t, s)lw(s)lds
to
(3.9.6)
Iw(s)I(<>+1)/a
H(t, s)
1/<>() ds.
a
s
Letting
A
and
(aH(t, s0/(0+1)
B
Iw(s)1
Iw(s)lh(t,s) - aH(t,s)
for all t
1
it
H(t, to)
to
+1
to,
Iw(s)I(<>+l)/Ct
a1/Ct(s)
Hence, (3.9.6) leads to
t, to
A= a
a o /(0+1) aO /(Ct+1) (s)hCt(t, s)

(a + 1)0 Ho'/(0+1)(t, s)
then from Lemma 3.2.1, we find for t > s
H(
a 1/(0+1)(s)
H(t, s)q(s)ds S; w(to) + H()

t, to
<
it
to
a(s)hCt+l(t, s)
aoH<> (
t,s) .
a(s)h<>+l(t s)
HCt( ') ds (3.9.7)
ao
t, s
to. Consequently, we obtain
it
to
[H(t,S)q(s) _ a(S)h Ct +1(t,S)] ds < w(to)

aoH<>(t,s)
for
~ to.
Chapter 3
170
Taking the upper limit as t ---+ 00, we get a contradiction, which completes
the proof.
A simpler version of Theorem 3.9.1 can be stated as follows.

Theorem 3.9.2.
Let the function H be as in Theorem 3.9.1 such
that (3.9.1) hold, and suppose that H has a continuous and nonnegative
partial derivative with respect to the second variable. Let h: V ---+ lR be
a continuous function with
- osH(t,s) = h(t,s)Ha/(o+l)(t,S)
If
limsup H( 1
t, to
t->oo
it
to
for
(t,s) E V.
(3.9.8)
[H(t, s)q(s) - a(s) ho+1(t, s)] ds

aD
(3.9.9)
00,

Next, we shall prove the following result which extends and improves
Theorem 3.9.1.
Theorem 3.9.3. Suppose the functions H and h are as in Theorem 3.9.1
such that conditions (3.9.1) and (3.9.2) hold. If there exists a nondecreasing
function p{t) E C1([to,00),lR+) such that
lim sup H( 1
t, to
t->oo
r [H(t'S)P(S)q(S)
lto
_ a(s)p(s) (h(t,s)
aoH"'(t, s)
+ p'es) H(t,S))"'+l] ds =
pes)
00,
(3.9.10)
Proof. Let x(t) be a nonoscillatory solution of equation (3.1.1). Without
loss of generality we assume that x(t) #- 0 for t ~ to. Now, we define
wet) = p(t)a(t)lxl(t)I"'-lx'(t)/(lx(t)I"'x(t)) for t ~ to. Then for every
s ~ to, we have
I
w (s) = - p(S)q(S)
p'es)
Iw(s)I(",+l)/a
peS) w(s) - a (a(s)p(s))l/o'
Multiplying (3.9.11) by H(t,s) for t

to t, we get
r H(t, s)w'(s)ds
lto
it
to and integrating from to
H(t, s)p(s)q(s)ds +
to
(3.9.11)
it -(-)
p'es)
to
Iw(s)l(a+l)/a
-a lto H(t, s) (a(s)p(s))l/a ds.
w(s)ds
l:
Thus, from (3.9.5), we find
~ H(t, to)w(to)+
H (t, s)p(s)q(s)ds
1:
(h(t, s)+
171
~g} H(t, s)) Iw(s)lds
Iw(s)l(o+l)/o
-a lto H(t, s) (a(s)p(s))1/0ds.
(3.9.12)
Therefore, in view of Lemma 3.2.1 with
(H(
= a
t, s
))<>/(0+1)
Iw(s)1
(a(s)p(s))1/(<>+l) '
and
a )<> ( a(s)p(s) )<>/(<>+1) (

B = ( a +1
(aH(t, s))<>
h(t, s)
for t> s
p'(s)
+ p(s) H(t, s)
)<>
to, we obtain
Iw(s)1 ( h(t, s)
p'(s))
Iw(s)I(<>+l)/<>
p(s) H(t, s) - aH(t, s) p1/<>(S)
a(s)p(s)
(
~ (a + 1)<>+lHo(t, s) h(t, s)
(3.9.13)
p'(s)
) <>+1
p(s) H(t, s)
From (3.9.12) and (3.9.13), we ha.ve
t r
1
a(s)p(s) (
H(t, to) lto lH(t, s)p(s)q(s) - aoH<>(t, s) h(t, s)
p'(s)
\ <>+1]
p(s) H(t, s))
ds
~
Now, taking the upper limit as t -t
00,
w(to).
(3.9.14)
we obtain a contradiction.
Corollary 3.9.1. Let condition (3.9.10) in Theorem 3.9.3 be replaced by

1
t~~ H(t, to)
it
to
a(s)p(s) (
p'(s)
)<>+1
H<>(t, s) h(t, s) + p(s) H(t, s)
ds <
and
limsup H( 1
it
)
H(t, s)p(s)q(s)ds
t, to to
then the conclusion of Theorem 3.9.3 holds.
t-J.oo
00
00,
(3.9.15)
(3.9.16)
Example 3.9.1. Consider the differentia.l equation
(t"lx'(t)1 0
1 x'(t))'
[-yfY- 3 (2-cost)
+ {Y-2 sint] Ix(t)I<>-lx(t)
= 0
(3.9.17)
172
Chapter 3
for t 2:: to > 0, where , is an arbitrary positive constant, a and

constants such that v < a - 2, a < 1.
1/
are
Here, we choose p(t) = t 2 and H(t, s) = (t - S)2 for t 2:: s 2:: to.
Then since p(t)q(t) = [fY(2 - cost)]', we find
p(s)q(s)ds 2:: fY - ko
for some constant ko > 0, and therefore,
It:
1 {t(
)2 ( ) ( )d
2fY
kl
k2
k
t2lt o t-s psqs s 2:: (!+I)(!+2)+t 2 +T- 0,
where
and
2t"+1
k2 = 2koto _ _0_.
,+1
Hence, condition (3.9.16) is satisfied. On the other hand, since

_1
t2
it
to
s 1/+2
(t - s) 20
= 2 +1t o - 1
0
2
)0+1 ds
2(t-s)+_(t-S)2
S
t ) 1-0 tl/- a +2
sl/-o+l(t _ s)I- 0 ds :::; 2 0 +1 1 _ J>.
tl/- o +2
0
l/-a+2
condition (3.9.15) is also satisfied. Thus, from Corollary 3.9.1 equation

Let H(t, s) = (t - s)n for t 2:: s 2:: to, where n> a is a constant.
Then by Theorem 3.9.3 the following corollaries are immediate.
Corollary 3.9.2. If there exists a nondecreasing function p E C 1([t o, 00),
]R+) such that
limsup
t ..... oo
~ ltor
[(t - s)np(s)q(s)
~o a(s)p(s)(t -
s)n-a-l
(~g? (t -
s)
n)
0+1] ds = 00,
(3.9.18)
Corollary 3.9.3. If
limsup
t-400
r [(t - stq(s) -
lto
n o+ 1 a(s)(t - s)n-o-l] ds
ao
00, (3.9.19)

Theorem 3.9.4. Let the functions H, hand p be as in Theorem 3.9.1
such that conditions (3.9.1) and (3.9.2) hold, and let
o<
inf
s?.to
{liminf
t-400
:?'
s))} :::; 00
t, to
(3.9.20)
173

and
1
li~~p H(t, to)
lto
a(s)p(s) (
H"'(t, s) h(t, s)
nE
Ifthere exists a function
li~~p
H(:, T)
)"'+1
ds < 00.
(3.9.21)
C([to, 00), lR) such that for every T 2:: to,
[H(t, s)p(s)q(s)
a(s)p(s) (
- ooH"'(t, s) h(t, s)
and
oo
to
where n+(t)
p'es)
+ pes) H(t, s)
(3.9.22)
p'es)
)"'+1]
pes) H(t, s)
ds >
(n+ (s) )(0+1)/<>

..:...,-..,...:,...:..:..,-,...,...,....,.-ds
(a(s)p(s1/'"
= max{O, n(t)},
00
neT)
'
(3.9.23)
Proof.

x(t) =I 0 for t 2:: to. Define the function wet) as in Theorem 3.9.3, to get
(3.9.12) and (3.9.14). Then for t 2:: T 2:: to, we have
li~~p
H(:, T)
[H(t, s)p(s)q(s)
_ a(s)p(s) (h(t,s)
+ p'es) H(t'S)"'+1] ds ::::; weT).
00
pes)
Thus, from condition (3.9.22) it follows that
neT) : : ;
and
lim sup H( 1
t, to
t ..... oo
it
to
weT)
for every
T 2:: to
H(t, s)p(s)q(s)ds 2:: n(to).
Now, we define functions
U(t)
H(t,to)
and
Wet) = H( 1 )
t, to
where
(J(t, s) = h(t, s)
Iw(s) 1(<>+1)/'"
lto H(t,s) (a(s)p(s)l/",ds
it
to
(J(t, s)lw(s)lds,
p'es)
pes) H(t, s),
t 2:: s 2:: to
(3.9.24)
(3.9.25)
Chapter 3
174
Then from (3.9.12) and (3.9.25), we find
< w(to) -limsup H( 1 )
liminf[U(t) - W(t)]
t-+oo
t, to
t-+oo
:s:
w(t o) - fl(to) <
1t
to
H(t, s)p(s)q(s)ds
00.
(3.9.26)
Next, we claim that
1=
to
Iw(s)l(a+l)/a
~'--C-'-....,...-;-:-:;-;-ds
(a( s )p(s) )I/a
<
00.
(3.9.27)
00.
(3.9.28)
Suppose to the contrary that
= Iw(s)l(a+l)/a
-:--:'--:--c-:-:-:;-;-
(a(s)p(s))l/a
to
ds
By condition (3.9.20), there exists a positive constant

inf
s2>:to
{ lim inf
t-+=
II(t,s) }
(
)
H t, to
>
such that
> O.
(3.9.29)
Let /-l be an arbitrary positive number. Then it follows from (3.9.28) that
there exists a tl > to such that
1
t
to
Iw(s)I(a+l)/a
/-l
( ) ( 8 )) l/a ds > -~
(asp
Therefore, for all t:::: t l
U(t)
= H(t,Q to)
1t
1t (a
(1
H(t, s)d
for all
IW(T)I(a+l)/a
(a(T)p(T))l/a dT
) (1
H(t, to)
t (-.!!-.H(t, 05)) (1
H(t, to) it,
as
Q
to
to
--H(t, s)
as
/-lQ
~H(t, to)
1t
t,
to
t:::: tl
to
S
to
IW(T)I(a+l)/a dT ) ds
(a(T)p(T))l/a
IW(T)I(a+l)/a dT) ds
(a(T)p(T))l/a
-'!!-'H(t. s)ds = /-lH(t, td.

as'
~H(t, to)
From (3.9.29), we find that there is a t2 > tl such that H(t, td/ H(t, to) ::::
for all t:::: t 2, which implies that U(t):::: /-lQ for all t:::: t2' Since jJ
is arbitrary
lim U(t) = 00.
(3.9.30)
~
t-+oo
Further, consider a sequence {Tn}~=l in (to,

and
00)
such that limn-+oo Tn
= 00
lim [U(Tn) - W(Tn)]
n--+oo
= liminf
[U(t) - W(t)] <
t---+oo
00.
Oscillation and nonoscillation of haH-linear differential equations
175
Then there exists a constant M such that

(3.9.31 )
for all sufficiently large n. Since (3.9.30) ensures that
lim U(Tn)
n-+oo
(3.9.31) implies that
lim W(Tn)
n-+oo
(3.9.32)
00,
(3.9.33)
00.
Combining (3.9.31) and (3.9.32), we derive for n sufficiently large
Therefore, we have
W(Tn)
1
U(Tn) > 2"
for all large
(3.9.34)
n.
Now from (3.9.33) and (3.9.33), we obtain
.
W<>+l(Tn)
hm
=
n-+oo U<>(Tn)
(3.9.35)
00.
On the other hand by Holder's inequality, we have
<
and hence
1
<>
H(T)
n, to
Tn
to
f3<>+l(T s)
a(s)p(s) H<>(T n, ) ds.
n, S
Chapter 3
176
Finally, it follows from (3.9.35) that

lim
n--+oo
H(T
to
n,
which gives
.
11m
t--+oo
H t, to
Tn
to
(3<>+I(T s)
a(s)p(s) HO(T n, ) ds
00,
n, S
a(05)p(o5)
to
(30+1(t,S)
( ) ds =
HO: t,05
00.
But, this contradicts condition (3.9.21) and hence (3.9.28) holds.
Now, from (3.9.24), we have

roo
.Ito
(Q+(S))(u+l)/o:
(a(s)p(s))l/O ds <
OO
to
!w(s)!(a+l)/o
(a(s)p(s))l/ods <
00,
Theorem 3.9.5. Let the functions H, hand p be as in Theorem 3.9.4

such that conditions (3.9.1), (3.9.2) and (3.9.20) hold, and
it
liminf H( 1 )
t--+oo
t, to
to
H(t, s)p(s)q(s)ds <
If there exists a function Q E C([to,
liminf
t--+oo
(1
H t, T
.IT
00),
00.
(3.9.36)
JR.) such that for every T:2: to,
[H(t, s)p(s)q(s)
a(s)p(s) (
- aoHo(t, s) h(t, s)
(3.9.37)
p'(s)
)"'+1]
p(s) H(t, s)
ds :2: Q(T)

x(t) f. 0 for t :2: to. As in Theorem 3.9.3, (3.9.12) and (3.9.14) are satisfied
and proceeding as in Theorem 3.9.4, (3.9.24) holds for t:2: T :2: to. Using
condition (3.9.36), we find
lim sup [U(t)-W(t)] ::; w(to)-liminf H( 1

t--+oo
t--+oo
t, to
it
to
H(t,05)p(o5)q(s)do5 <
00,
where U(t) and W(t) are as in the proof of Theorem 3.9.4. Now, it
follows from condition (3.9.37) that
Q(to) ::; liminf H( 1 )

t--+oo
..
-11mlllf
t--+oo
t, to
it
it
1
()
H t, to
to
H(t, s)p(s)q(s)ds
to
a(s)p(s) (
p'(s)
)<>+1
H ( ) h(t,s)+-(-)H(t,s)
ds,
t, s
Ps
ao
177
so that condition (3.9.36) implies that

.,
hmmf
t--'too
ao
H( t, to )
1t a(s)p(s) (
H a (t, S ) h(t, s)
to
p'(s)
+ -(-)
P s H(t, s)
)a+1
ds < 00.
Let {Tn };;o=1 be a sequence in (to, 00) with limn--'too Tn = 00 such

that limn--'too [U(Tn) - W(Tn)] = limsuPHoo [U(t) - W(t)]. Then using
the procedure as in the proof of Theorem 3.9.4 we conclude that (3.9.27)
is satisfied. The remainder of the proof proceeds as in Theorem 3.9.4 and
hence omitted.
Now, define H(t, s) = (1r(t) -1r(s))n for t 2:: s 2:: to, where n> a is
a constant and 1r(t) E C 1([to,oo),lRo) with 1r'(t) = 1/a 1/ a (t) for t 2:: to
and limt--'too 1r(t) = 00. Then, H(t, s) is continuous on V = {(t, s) ; t 2::
s 2:: to} and satisfies H(t, t) = 0 for t 2:: to, H(t, s) > 0 for t> s 2:: to
Moreover, H(t, s) has a continuous and nonpositive partial derivative on
V with respect to the second variable. Furthermore, the function
h(t, s) = a1/:(t) (1r(t) _1r(s))(n-<>-1 l /(a+1 l ,
t > s 2:: to
is continuous and satisfies
= h(t, s)H<>/(<>+1l(t, s) for t> s 2:: to
- :s H(t, s)
We find that (3.9.20) holds, since for every s 2:: to,

1.
.
H(t, s)
11m
t--'too H(t, to)
If p(t)
1m
Hoc
(1r(t) _1r(s))n
(1r(t) -1r(tO))n
1.
= 1, then since
--1
1
limsup
a(s) _n_(1r(t)_1r(s))(n-<>-1 l /(<>+1 l
t--'tCXl 1rn(t) to
a1/"(s)
1t
<>+1
ds
1
lim -nn<>+la- 1/a(s)(1r(t) -1r(s)t- a - 1ds
t--'too 1r (t) to
n a +1 (1r(t) - 1r(to))n-<>
lim - - 0
t--'tCXl 1rn(t)
n_a
-,
condition (3.9.21) is satisfied. Thus, Theorem 3.9.4 leads to the following

corollary.
Corollary 3.9.4. Let n> a be a constant, and let 1r(t) E C1([to,00),lRo)

with 1r'(t) = a- 1/<>(t) and limHoo 1r(t) = 00. Suppose there exists a
function n(t) E C([to,oo),lR) such that
1
t
to
(n+(s))(<>+1l/a
l/a()
s
ds - 00
(3.9.38)
Chapter 3
178
and
li~~p
7r n1(t)
[7r(t) - 7r(s)tq(s)ds
n(T),
for all T
~ to
(3.9.39)
where n+(t) = max{n(t), O}. Then equation (3.1.1) is oscillatory.

In the case when H(t, s) = (t - s)n for t
we have the following corollary.
to and some n
> a,
Corollary 3.9.5. Let n > a be a constant and suppose there exists a

function n(t) E C([to, 00), IR) such that condition (3.9.38) holds. If
limsup ~
t-+oo t
for all T
rt (t _
JT
s)n-o-l [(t _ S)o+lq(S) _ nn+l a(s)] ds
ao
n(T)
(3.9.40)
to, then equation (3.1.1) is oscillatory.

(tVlx'(t)10-1x'(t))' +(tAcost)lx(t)IO-lX(t)
0 for t ~ to > 0, (3.9.41)
:s:
where v, .x, a are constants such that - 1 < .x 1, v < n, a -=1= 2 and
a 2.x ~ (a + l)(v - a). Taking H(t, s) = (t - s)2 for t ~ s ~ to, we find
1 l.t SV(t - s?-ods

2"
t to
<
tV (t - to)2-o
v>O
t 2 2-a
to (t - to)2-0
v<O
t 2 2-a
tV-a (
to ) 2-0
- - 1-v>O
2-a
t
'
{
{
~~
2 - a to
(1- to )2-0
t
'
< O.
Therefore, the condition limsuPt-+oo(1/t 2) ftto a(s)(t - s)n-o-lds < 00 is

satisfied, and for arbitrary small constant f > 0 there exists a h ~ to
such that for T ~ t 1,
lim sup -1
t-+oo t 2
it [
T
(t - s)2 SA cos S
SV (t - S)1-0] ds ~ - TA cos T (1 + a)1+ o
f.
Now, set n(t) = _TA cosT - f. Then there exists an integer N such
that (2N + 1)7r - (7r/4) > tl and if n 2': N, (2n + 1)7r - (7r/4) ::; T ::;
(2n+ 1)7r+ (7r/4), n(T) ~ oTA, where 0 is a small constant. Now, since

0. 2
>.
~ (a
00
to
+ 1)(1/ -
a), we obtain
(n+(s))(o+1)/o
1/0( )
179
ds > ~
5(0+1)/0 j(2n+1)7r+(7r/4)
L..n=N
(2n+l)7r-(7r/4)
1(2n+l)7r+(7r/4)
00
5(0+1)/0
n=N
(2n+l)7r-(7r/4)
sA(o+I)/o-(v/o)ds
ds
= 00.
Hence, all conditions of Corollary 3.9.5 are satisfied and therefore equation
3.9.2. Interval Criteria for the Oscillation

of Equation (3.1.1)
It is well-known that equations (2.1.1) and (3.1.1) have several similar
properties. For example, Sturmian comparison and separation theorems
for equation (2.1.1) have been extended to (3.1.1) in Section 3.2. Thus, the
zeros of two linearly independent solutions of (3.1.1) separate each other and
all nontrivial solutions of (3.1.1) are either oscillatory or nonoscillatory. We
note that most of the oscillation criteria obtained so far involve the integral
of the function q(t), and hence require the information of q(t) on the
entire half-line [to, (0). These results cannot be applied to the case when
q(t) has a 'bad' behavior on a large part of [to, (0), for example, when
ft"; q(s)ds = -00. However, from Sturm separation theorem, we note that
oscillation is only an interval property, i.e., if there exists a sequence of
subintervals [a;, b;] of [to, (0), ai -+ 00 such that for each 'i there exists
a solution of (3.1.1) that has at least two zeros in [a;, b;], then every
solution of (3.1.1) is oscillatory, no matter how 'bad' equation (3.1.1) is (or
a(t) and q(t) are) on the remaining parts of [to, (0). In this section we
shall use this idea to provide several oscillation criteria for (3.1.1). These
results in particular extend and improve those given in Section 2.5 for the
equation (2.1.1).
In what follows we say that the function H = H(t, s) belongs to the

class r, denoted by HEr if HE C(D, JR+) where D = {(t, s) : -00 <
S :::; t < oo} which satisfies condition (3.9.1), and has partial derivatives
8H/8t and 8H/8s on D such that
8
8tH(t,s) = hl(t,S)
(3.9.42)
and
where hI, h2 are continuous and nonnegative on D, hI, h2 E CLoc(D, JR).

Again, we shall let by 0.0 = (a
need the following two lemmas.
Lemma 3.9.1.
+ 1)0+1.
To prove our main results we
Assume that x(t) is a solution of equation (3.9.1) such
180
Chapter 3
that x(t) > 0 on [e, b). Let
w(t) = p(t)
where p(t)
HEr,
a( t) lx' (t) la-1 x' (t)

Ix(t)la- 1x(t) ,
(3.9.43)
C 1 ([to, 00), JR+) is any nondecreasing function. Then for any
b
Ib a(s)p(s)
I c H(b, s)p(s)q(s)ds :::; H(b, e)w(e) + c ooHa(b, s)
p'(s)
x [H(b,s) p(s)
+ h2(b,s)
]a+1
(3.9.44)
ds.
Proof. From equation (3.1.1) and (3.9.43), we have for s E [e, b),
p(s)q(s) =
,p'(s)
Iw(s)IC<>+1)/a
w (s) + p(s) w(s) - a (a(s)p(s)l/a
Multiplying (3.9.45) by H(b, s),

(3.9.1) and (3.9.42), we obtain
lbH(b, s)p(s)q(s)ds
(3.9.45)
integrating from e to b and using
-lb H(b, s)w'(s)ds + lb H(b, s)~(~j w(s)ds
-0
b
Iw(s)ICa+1)/a
I c H(b, c) (a(s)p(s)l/a ds
:::; H(b, e)w(e)
+ lb [H(b, s)~gj +h2(b, s)]
lw(s)lds
b
Iw(s)ICa+1)!a
-a I c H(b, s) (a(s)p(s)l/a ds.
Thus, from Lemma 3.2.1 with
_ (
A and
B
oH(b, s
(_0_)
0+1
a (
a!Ca+1)
Iw(s)1
(a(s)p(s)l/Ca+1)'
A-
+1
a(s)p(s) )a/ ca+1) (h 2(b, s) + H(b, S)p'(s)a

(oH(b, sa
p(s)
we get for t> s 2': to,
p'(s)
Iw(s)I Ca+1)/a
Iw(s)1 ( h2(b, s) + p(s) H(b, s) - aH(b, s) (a(s)p(s)l/a
<
-
a(s)p(s) (h (b s)
ooHa(b,s)
2,
+ p'(s) H(b
p(S)
S)a+1
181
and hence, we have
lb
H(b, s)p(s)q(s)ds
H(b ) ,()
,c u c
S;
( h2(b, s)
+ 00
Jb HI>
a(s)p(s)
(b )
,s
)00+1
p'(s)
p(s) H(b, s)
ds.
Lemma 3.9.2. Assume that x(t) is a solution of equation (3.1.1) such

that x(t) > 0 on (a, el. Let p(t) E C 1 ([to, 00), JR+) be a nondecreasing
function and u,(t) be as in (3.9.43) on (a, cl. Then for any HEr,
1
C
H(s, a)p(s)q(s)ds S; -H(c,a)'lL'(c)

x
1
c
a(s)p(s)
00
DoH (s, a)
p'(s)
Q+1
H(s,a)p(s)+h 1(s,a)
ds.
(3.9.46)
Proof.
Following Lemma 3.9.1, we multiply (3.9.45) by H(s, a)
integrate from a to c, to obtain
1H
(s,a)p(s)q(s)ds
=-1"
H(s,a)'lL"(s)ds+
-0
S;
Ja
1H(s,a):(~}'lL'(S)dS
c
1'lL'(s) 1(00+1)/",
H(s, a) (a(s)p(s))l/Q ds
r[
p'(s)
1
-H(c, a)'lL'(c) + Ja H(s,a) p(s) +h1(s,a)j l'lL'(s)lds
-0
r H(s, a) Iw(s)I(",+l)/oo
(a(s)p(s))l/OO ds.
Ja
Thus, from Lemma 3.2.1 with
A = (H(
))"'/(Q+1)
Iw(s)1
o
s,a
(a(s)p(s))l/(ct+ 1) '
and
we get for t > s :::: to,
1'lL'(s) 1( h1(s,a)
S;
p'(s))
Iw(s)I(<>+l)/<>
p(s) H(s,a) - oH(s,a) (a(8)p(s))1/<>
a(s)p(s) (
p'(s)
)<>+1
H (s, a ) h1(s,a) +-(-)
H(s,a)
p s
00
and
182
Chapter 3
and hence, we have
1
C
H(s, a)p(s)q(s)ds
<::::
-H(c, a)w(c)
x ( hds,a)
+ -ao
1
c
a(s)p(s)
HQ(
)
S,
p'(s)
)0+1
p(s) H(s,a)
ds.
Corollary 3.9.6. Let x(t) be a solution of equation (3.1.1), x(t) > 0

for t 2:: T 2:: to and let w(t) and p(t) be as in Lemma 3.9.1. Then for
any HEr and a, b, c E lR such that T <:::: a < c < b,
H(~,a)
<::::
1
c
H(s,a)p(s)q(s)ds+
1
c
-(-)
H c, a
+ H(b, c)
H(~,c)
lb
H(b,s)p(s)q(s)ds
a(s)p(s) (
p'(s)
)0+1
Q() H(s,a)-(-) +h1(s,a)
ds
aoH s, a
ps
lb
c
a(s)p(s) (
p'(s)
aoHCb, s) H(b, s) p(s)
+ h2(b, s)
)0+1
ds.
(3.9.47)
Proof. Divide (3.9.44) and (3.9.46) by H(b, c) and H(c, a), respectively
and add them together.
Corollary 3.9.7. Assume that p(t) E C 1([to, 00), lR+), p'(t) 2:: 0 for
t 2:: to and for some c E (a, b) and HEr,
1
H(c, a)
Jar
H(s, a)p(s)q(s)ds + H(b, c)
> H(c,a)
Ib
c
H(b, s)p(s)q(s)ds
r aoH<>(s,
a(s)p(s) (
p'(s)
)Q+1
a) H(s,a) p(s) +h1(s,a)
ds
Ja
lb
a(s)p(s) (
p'(s)
)<>+1
aoHo(b,s) H(b,s) p(s) +h2(b,s)
ds.
(3.9.48)
Then every solution of equation (3.1.1) has at least one zero in (a, b).
1
+ H(b, c)
Proof. Assume to the contrary that x(t) is an eventually positive solution

of equation (3.1.1) on (a, b). From Corollary 3.9.6, there exists a T 2:: to
such that inequality (3.9.47) holds for any HEr and a, b, c E lR
satisfying T <:::: a < c < b. This contradicts (3.9.48) and completes the
proof.
Now, we state and prove the main interval oscillation criterion for
(3.1.1).
183
Theorem 3.9.6. If for each T ~ to there exist HEr, p E Cl([to, 00),

1R+), p'(t) ~ 0 for t ~ to, and a, b, c E 1R such that T:::; a < c < b
Proof. Pick a sequence {Ti} C [to, 00) such that Ti -t 00 as i -t 00.
By the hypotheses for each i E 1N there exist ai, bi , Ci E lR such that
T;, :::; ai < Ci < bi and (3.9.48) holds with a, b, c replaced by ai, bi , Ci,
respectively. From Corollary 3.9.7 every solution x(t) has at least one
zero, ti E (ai,b i ). Noting that Ti :::; ai < ti, i E 1N we find that every
solution has arbitrarily large zeros. Thus, every solution of equation (3.1.1)
is oscillatory.
Theorem 3.9.7. If there exist HEr and p E C 1 ([to, oo),lR+), p'(t) ~ 0

for t ~ to such that for any r ~ to,
J: H(s, r)p(s)q(s)ds
lim sup
00
(3.9.49)
00,
(3.9.50)
t-+oo
and
lim sup
t-+oo
1: H(t,s)p(s)q(s)ds
Ir
a s)p(s (
W. t,s)
H(t, s)
p'(S)
p(s)
+ h2(t, s)
)<>+1
ds

Proof. In (3.9.49) let r = a = T
that
H(s,a)p(s)q(s)ds > ao
In (3.9.50), we choose r
JbH(b, s)p(s)q(s)ds
C
to. Then there exists a c > a such
jC HO:(s,a)
a(s)p(s) (
p'(s)
)<>+1
H(8,a) p(s) +h1(s,a)
ds.
= c.
> ao
(3.9.51 )
Then there exists a b> c such that
Jb HO:(b,
a(s)p(s) (
p'(s)
)0:+1
s) H(b, s) p(s) + h2(b, s)
ds.
C
(3.9.52)
Combining (3.9.51) and (3.9.52), we obtain (3.9.48). Now, it follows from
Theorem 3.9.6 that equation (3.1.1) is oscillatory.
The following result is equivalent to that of Theorem 3.9.6.

Theorem 3.9.8.
If there exist HEr and a nondecreasing function
184
Chapter 3
p(t)
C 1 ([to, 00), JR+) such that for any r 2 to,
lim sup
t-400
jt LfH(s, r)p(s)q(s) _
r
a~~(s))
Qo
(H(S, r) p'((s))
s, r
p s
+ hI (s, r)\,,'+I] ds
)
> 0
(3.9.53)
and
limsuPjt fH(t, s)p(s)q(s)r
t-400
a~~(s))
s, r
Qo
(H(t, s)p'((s))
p s
+ h2(t, S)\,,'+1] ds
)
> 0,
(3.9.54)
Proof. In (3.9.53) let r = a = T 2 to. Then there exists a c> a such

that
1
c
a(s)p(s) (
p'(s)
H(s,a)p(s)q(S) - QoH"'(s,a) H(s,a) p(S)
In (3.9.54), we choose r
+ h1(s,a) )"'+1]
ds > o.
(3.9.55)
= c. Then there exists a b> c such that
a(s)p(s) (
p'(s)
H(b, s)p(s)q(s) - QoH"'(b, s) H(b, s) p(s)
+ h2(b, s) )"'+1]
ds >
o.
(3.9.56)
Combining (3.9.55) and (3.9.56), we obtain (3.9.48). Now, the conclusion
The following result is a special case of Theorem 3.9.6. In fact, we let

= 1 to obtain
p(t)
Corollary 3.9.8. If for each T 2 to there exist HEr and a, b, c E JR

such that T:::; a < c < band
1 a)
H(c,
>
1
_1_1c
c
1
H(s,a)q(s)ds+ H(b,c}
H(c, a)
a(s)hf+1(s,a) ds
QoH"'(s, a)
lb
+ _1_1b
c
H(b,s)q(s)ds
H(b, c)
a(s)h~+I(b,s) ds
QoH"'(b, s)
,
(3.9.57)

If condition (3.9.1) holds, and (3.9.42) is replaced by
:tH(t, s)
= h 1(t, s)H",/(a.+1)(t, s)
and :s (t, s)
= -h2(t, s)Ha./(n+l)(t, s),

(3.9.58)
Oscillation and nonoscillation of half-linear differentia'] equations
185
where hI, h2 are nonnegative and continuous functions on V, hI, h2 E

.c Lac (V, lR) , then we shall say H E f*. In this case Corollary 3.9.8 is
reduced to the following result.
Corollary 3.9.9. If for each T
such that T:S: a < c < band
1 a)
H(c,
>
H(s, a)q(s)ds
:0 [H(~,a) l
to there exist H E f* and a, b, c E lR
+ H(b,1 c)
lb
c
H(b, s)q(s)ds
a(s)hr+l(s,a)ds+
H(~,C) Ib a(S)h~+I(b'S)dSl'
(3.9.59)

Next, we state the following interesting result.
Corollary 3.9.10. If for each T ~ to there exist H E f and a, b, c E lR
such that T:S: a < c < b, and either
c
J a H(s,0,)q(s)d8
or
lb
H(b, s)q(S)d8
> _1 JCa(S)hr+l(s,a)dS
aD
Hc>(s, a)
lb
> _1
ao
a(s)h~+I(b,s)d8
HO'(b, 8)
,
(3.9.60)
(3.9.61 )

In the case when condition (3.9.58) holds Corollary 3.9.10 reduces to
Corollary 3.9.11. If for each T ~ to there exist HE f* and a, b, c E
lR such that T:S: a < c < b, and either
or
Ib
aD
lb a(s)h~+I(b,
H(s, a)q(s)ds > -1

H(b, s)q(s)ds >
aD
a(s)hr+l(s, a)ds,
(3.9.62)
s)ds,
(3.9.63)

(a(t)lx'(t)IO'-IX'(t))'
+ q(t)lx(t)Ic>-lx(t)
0,
where a > 0 is a real constant,
a(t) =
e-2n(0'+1)
for
for
2n < t < 2n + 1
2n + 1 < t < 2n + 2
(3.9.64)
186
Chapter 3
and
q(t) =
2n < t < 2n + 1
for
et-2n(a+2)
2n + 1 < t < 2n + 2,
for
f;
where q1 > 0 is a constant and n E:IN. Obviously,

q(s)ds is
convergent as t --+ 00. For T ~ 0 there exists n E:IN such that
2n ~ T. Let a = 2n, b = 2n + 1 and H(t,s) = (e t - eS )<>+1. Then,
HE f*, h1(t,S) = (a + I)e t and h2(t,S) = (a + I)e s . To show that
equation (3.9.64) is oscillatory, by Corollary 3.9.11 it suffices to verify that
for all c E (a, b) either condition (3.9.62), or (3.9.63) holds. In fact,
by a simple computation, we find that condition (3.9.62) holds for each
I': = C - 2n E (0,1) if
(e E _I)Q+2
a+2
q1 (eE(<>+l) _ 1) > a + 1
and condition (3.9.63) holds for each
q1
= 2n + 1 - c E (0,1) if
1':1
a+2
(e_e E1 )<>+2
(e -
(3.9.65)
> a + l'
eE1 (<>+1)
(3.9.66)
For an appropriate choice of q1, we note that both the conditions (3.9.65)
and (3.9.66) hold, and hence equation (3.9.64) is oscillatory.
For the case
denote them by
is denoted by f
H = H(t - s) E f*, we have h 1(t - s) = h2(t - s) and

h(t - s). The subclass of f* containing such H(t - s)
o.
Applying Corollary 3.9.9 to fo, we get the following result.

Theorem 3.9.9. If for each T
such that T ~ a < c, and
to there exist HE fo and a, c E 1R
H(s, a)[q(s) + q(2c -- s)Jds > -1

ao
h<>+l(s, a)[a(s) + a(2c - s)Jds,

(3.9.67)

Proof. Let b = 2c - a. Then, H(b - c) = H(c - a) = H((b - a)/2)
and for any y E .e[a, b], we have
y( s )ds =
y(2c - s )ds. Hence,
f:
I:
f: H(b - s)q(s)ds f: H(s - a)q(2c - s)ds

=
Ib
a(s)h<>+1(b - s)ds
Ib
and
a(2c - S)h""+l(S - a)ds.
Thus, (3.9.67) implies that (3.9.59) holds for each HE fo, and therefore,
187
(lx'(t)IO-1x'(t))'
+ q(t)lx(t)IO-1X(t)
0,
(3.9.68)
where a > 0 is a constant, and
5(t - 2n) for 3n::; t ::; 3n + 1

{ 5( -t + 3n + 2) for 3n + 1 < t ::; 3n + 2
-n for 3n + 2 < t < 3n + 3, n E lN o.
q(t) =
For any T?: 0 there exists n E lN o such that 3n?: T. Let a =

3n, c=3n+1, and H(t-s)=(t-S)o+l. Then, h(t-s)=a+1. It
is easy to verify that condition (3.9.67) holds, and hence equation (3.9.68)
is oscillatory by Theorem 3.9.9. Note that in equation (3.9.68), we have
IoDO q(s)ds
-00.
Next, we choose H(t - s) = (t - s)n for n > a. Then, HEro and

h(t - s) = n(t - s)(n-a-l)/(a+!). Then, we have the following oscillation
criterion for (3.1.1).
Theorem 3.9.10. Equation (3.1.1) is oscillatory if for some n > a and
for any r?: to either
the following inequalities hold
(I)
li~~p
1t
[(s - r)nq(s) -
limsup
1t
[(t - s)nq(s) - nO+! (t - s)n-a-la(s)] ds > 0,
and
t--+oo
n::l (s - r)n-a-la(s)] ds >
aD
or
(II) the following inequality holds
lim sup
t--+DO
1~
t
n o + 1 (s_r)n-o-l (a(s)+a(2t-s))] ds
(s-r)n(q(s)+q(2t-s)) - __
aD
> O.
The following corollary is an immediate consequence of Theorem 3.9.10.
Corollary 3.9.12. Let a(t) = 1. Then equation (3.1.1) is oscillatory if
for any r?: to and some n > a either
(I)
the following two inequalities hold

1
limsup tn-a
t--+DO
It
r
(s - r)nq(s)ds >
ao(n - a)
Chapter 3
188
and
or
(II) the following inequality holds
f:
a- I /", (s)ds for t 2: r 2: to

For our next result, we define R(t) =
and let H(t, s) = (R(t) - R(s))n for t 2: s 2: to, where n > 0 is a
real constant. We let hI and h2 be defined in such a way that condition
(3.9.58) holds. In fact, we have
hl(t, s) = na-I/"'(t)(R(t) - R(s))(n-",-I)/(",+I)
and
h 2 (t, s)
= na-I/"'(s)(R(t) - R(s))(n-"'-I)/(a+I).
Corollary 3.9.13. Equation (3.1.1) is oscillatory if lim R(t) = 00 for

t-4oo
some n > 0, and for each r 2: to the following two inequalities hold
and
li~~p Rn-"'(t)
It
r
n",+1
(R(t) - R(s)tq(s)ds > oo(n - 0)'
Proof. The proof is easy and hence omitted.

(t"'lx'(t)I"'-lX'(t))'
+c'"I Ix(t)I"'-IX(t)
tnt
0,
(3.9.69)
a- I /", (s)ds =
where 0 > 0 and C > 0 are constants. Let R(t) =
Int, t > 1. Then, we have H(t, s) = (R(t) - R(s))n = (In(t/s))n for
t 2: s > 1, where n > 0 is a constant. It is easy to check that all the
conditions of Corollary 3.9.13 are satisfied and hence equation (3.9.69) is
oscillatory for all c > 0 and 0 > O.
3.9.3. The Weighted-Average Oscillation Criteria

Here, we shall extend Lemmas 2.5.1 and 2.5.2, and Theorems 2.5.4 and
2.5.5 to equation (3.1.1). For this, let n be the set of all nonnegative
189
locally integrable functions g on [to, 00), to :2: 0 satisfying the condition
li~~p (
(l/Ct)-k
g(s)ds
{Gk(oo) - Gk(t)} > 0 for some k E [O,l/a),

(3.9.70)
where
Gk(t)
ft g(s)
(f g(r)dr)k
(f a(r)g<>+1(r)dr)
ds.
(3.9.71)
/Ct
If Gk(oo) = 00 in (3.9.70), then gEn. Let no be the set of all

nonnegative locally integrable function g on [to, 00) satisfying
. t
a(s)gCt+1(s)ds
t
J g( s )ds
hm
t---+oo
(3.9.72)
O.
It is clear that for (3.9.70) or (3.9.72) to be satisfied by a nonnegative

function g it is necessary that
f= g(s)ds
00.
(3.9.73)
On the other hand every bounded nonnegative locally integrable function

g satisfying (3.9.73) belongs to no and no c n. Further, since all
nonnegative polynomials are members of no, no contains some unbounded
functions. Members of the classes n and no will be called weight
functions.
If (3.1.1) is nonoscillatory and x(t) is its nontrivial solution, then there
exist a number T:2: to and a function w(t) = a(t)'l/J(x'(t))N(x(t)) E
C1([T, 00), IR), which satisfies
w'(t)
t:2: T.
q(t) - aa- 1 /<>(t)lw(t)I(Ct+1)/Ct,
(3.9.74)
Clearly, (3.9.74) is equivalent to the integral equation
w(t) = w(s) - a
it
a- 1 / Ct (r)lw(r)I(<>+1)/ Ct dr
-it
q(r)dr
(3.9.75)
for t:2: s :2: T. In what follows for g E n, we shall denote by
()
A s t
9
g(r)
r q(u)dudr
J: g(r)dr
;:.;S,,---,-,-~s---'---'---
Lemma 3.9.3. Assume that w(t) satisfies equation (3.9.74) on [T, 00)
for some T:2: to. If there exists g E n such that
liminf A g (., t)
t--+oo
> - 00,
(3.9.76)
Chapter 3
190
then
IX) a-l/O(s)lw(s)l(o+l)/Ods <
Proof. Let A(s, t)
00.
= Ag(s, t) and assume that
Joo a-l/O(s)lw(s)l(o+l)/ods
(3.9.77)
00.
Multiplying both sides of equation (3.9.75) by g(t) and integrating it from

~ to t, we obtain
it
=
g(s)w(s)ds
w(~)ltg(S)dS - ltg (S)1 8q (T)dTdS -
= w(~)ltg(S)dS =
A(~, t)l tg (S)dS -
[w(~) - A(~, t)]l tg (S)dS -
where t
al tg (s) 18a-!;(T)lw(T)IQ!, dTds
al tg (s) lta-!;(T)IW(T)I "'-::' dTds
al tg (s)
1
8
a-!; (T)lw(T)1 "'!' dTds,

(3.9.78)
~ ~ ~
w(~)
Since 9 E
T. From equation (3.9.75), we have
w(T) -
n,
A(~,t)
if.
q(s)ds - a hf. a- l / O(s)lw(s)l(o+1)/ods.
(3.9.73) holds. This implies
I~g(s)dsA(T,t)_
If. g(s)ds
t~ g(s)ds A(T, t) If. g(s)ds
(f.q(s)ds_I';g(s)/;q(T)dTds
If. g(s)ds
iT
if.
T
q(s)ds + 0(1)
as
t~
00.
Thus, we get
w(~)
A(~,
w(T) _
t)
I~ g( s )ds A(T, t)
If. g(s)ds
-a hE a- l / O(s)lw(s)l(o+1)/Ods
Since 9 E
>.-1/0
n
<
there exists a positive number
(1)
- a
limsup
>.
(Jt g(s)ds )
+ 0(1)
as t
> 0 such that
(l/o)-k
~ 00.
(3.9.79)
[Gk(oo) - Gk(t)],
t-400
(3.9.80)
191
where k is as in (3.9.70). It follows from (3.9.76), (3.9.77) and (3.9.79)

that there exist two numbers tl and t2 with t2 ~ tl ~ T such that
(3.9.81)
Let yet) =
Jtt, g(s)w(s)ds.
Then Holder's inequality for s ~ tl yields
ly(s)I(O+l)/O:S; ([: a-l/C>(T)lw(T)I(O+l)!OdT)
([,8 a(T)gO+l(T)dT)
I/O.
Thus, from (3.9.78) and (3.9.81), we find
yet) :s; -A tg(s)ds-a tg(s)ly(s)1 n;;' (ra(T)90+1(T)dT) -!.dS = -pet).
it,
it,
it,
(3.9.82)
Hence,
P'(t)
Ag(t)
+ ag(t)ly(t)l(o+l)/<>
and
o :s;
t g(s)ds
it,
1,
a(s)go+1(s)ds
)-1/0
:s; pet) :s; ly(t)l.
(3.9.83)
(3.9.84)
Now, from (3.9.82), (3.8.83) and (3.9.84) it follows that
P'(t)pk-l-(l/<(t) > P'(t)pk(t)ly(t)I-(<>+l)/<>
> aAkg(t)
(I:
9(S)dS) k
(I:
a(s)g<>+1(S)dS) -1/<>.
For k < l/a, we integrate the above inequality from t

let u ~ 00, to obtain
t2 to u and
Therefore, from (3.9.84), we get
A-1/<>
-a- ~
; -
) ( t
kit! g(s)ds
) (l/<-k
[Gk(OO) - Gk(t)],
Lemma 3.9.4. Assume that wet) satisfies equation (3.9.74) on [T,oo)

for some T ~ to. If
a-l!<>(s)lw(s)I(<>+l)/<>ds < 00, then for any
9 E no, limt--+oo A g (., t) exists.
Joo
192
Chapter 3
Proof. As in Lemma 3.9.3, (3.9.78) holds. This implies that
A(C t) - w(C) ."

-.,
Since 9 E
-c'--_=---_.,.--_ _ _ _ __
I; g(s)ds
I; g(s)ds
(3.9.85)
no,
(3.9.73) holds. Thus, we have
rt
lim
Hoo
t g(s)w(s)ds - 0: t g(s).r a- 1/O(r)lw(r)I(O+1)/odrds
fS
0+1
J g(s)J a-"Q(r)lw(r)l--;>drds
I g(s )ds
00
n+1
a-"Q(s)lw(s)l--;>ds <
00.
By Holder's inequality, we find
o<
lim
- Hoo
<
II; g(s)w(s)dsl
-'-----------:;:-t- - - - - - ' -
11m
I g(s)ds
1, t a(s)g<>+l(s)ds)
1/(<>+1) (
1, t a- 1/<>(s)lw(s)li<>+lJ/<>ds)
<>/(<>+1)
I g(s)ds
- Hoo
Hence, in view of (3.9.85), limt--+oo Ag(~, t) exists and
From Lemmas 3.9.3 and 3.9.4 the following oscillation criterion for
(3.1.1) is immediate.
Theorem 3.9.11. If there exists 9 E
lim Gk(t) =
t-HX)
00
or, there exists 9 E

-00
for some
no
such that
[0,1/0:)
and
lim A g (., t) =
t~oo
00,
(3.9.86)
such that
< liminf A g (., t) < limsup A g (., t) <

t--+oo
t--+oo
00,
(3.9.87)

Corollary 3.9.14. Suppose g(t) and h(t) are two nonnegative bounded
functions on [T,oo) for some T;::: to with 100 g(s)ds = 00 = 100 h(s)ds.
If the functions a(t)g<>(t) and a(t)h<>(t) are bounded and
lim Ag(T, t) <
t--+oo
lim Ah(T, t),
t--+oo
(3.9.88)
Oscillation a.nd nonoscillation of half-linear differential equations
193
Proof. Let a and b be two numbers satisfying limt--+oo Ag(T, t) < a <
b<limt--+ooAh(T,t). Let f(t)=h(t) for T:St<tl, where tl is
such that Ah(T, tl) ~ band J; h(s)ds ~ 1. Further, let f(t) = g(t) for
tl :S t < t2, where t2 is such that Af(T, t2) :S a and J;' f(s)ds ~ 2.
This is possible because
J;' f(s) J; q(u)duds
J;' [h(s) - g(s)]f; q(u)duds
J;2 f(s)ds
J;' h(s)ds + Jtt,2 g(s)ds
J;2 g(s) J; q(u)duds
J;' g(s)ds
J;2 g(s)ds
J;' h(s)ds + Jtt,' g(s)ds
Ag(T, t2)[1
+ o(l)J + 0(1)
as
t2 ---+
00.
Continuing in this manner, we obtain nonnegative, bounded and nonintegrable function f(t) defined on [T,oo) such that limsuPt--+oo Af(T, t) ~
b > a ~ liminft--+oo Af(T, t) and
This implies that f(t) "= no and lim Af(T, t) does not exist. By Theorem
3.9.11 equation (3.1.1) is then oscillatory.
k(t) =
it
a-1/Q(s)ds
-t
00
as
t ---+
(3.9.89)
00.
to
If
-00
lItIs
It IS
< liE.~f k(t)

< lim sup
t--+oo
a-1/Q(u)q(u)duds
(3.9.90)
a-1/Q(u)q(u)duds :::;
00,

Proof. Let g(t) = a-1/Q(s). Then, g(t) E no and it follows from
(3.9.90) that limt--+oo A g (., t) does not exist. By Theorem 3.9.11 equation
(3.1.1) is then oscillatory.
Lemma 3.9.5. Assume that c(s) and P(s, t) are nonnegative continuous
functions for T:S s, t < 00. If
00
P(s, t)cCQ+1)/Q(s)ds :::; (a + 1)-CQ+l)/O:c(t)
for
~T
(3.9.91)
194
Chapter 3
then the equation
y(t)
c(t)
+0
IX) P(s, t)ly(s)I(<>+l)/nds
for
t:::: T
(3.9.92)
has a continuous solution y(t).
Proof. Let Yl (t) = c( t) and define

Yl (t)
S Y2(t)
S
IX) P(s, t)c(n+l)/a(s)ds
c(t)
+0
c(t)
+ 0(0 + 1)-(a+l)/oc(t) S
+ l)c(t)
(0
and Yl(t) S Y2(t) S (o+l)c(t) for t:::: T. Suppose Y1(t) S Y2(t) s ... ::;
Yn(t) ::; (0 + l)c(t), t:::: T where n is a positive integer. Then, we have
Yn(t)
c(t)
+0
c(t) +01
c(t)
00
P(s, t)IYn_1(S)I(0+1)/Ods
00
P(s,t)IYn(s)l(o+l) jo ds
+ 0(0 + 1)(Q+1)/0
00
Yn+1(t)
P(s, t)c(Q+1)jQ(s)ds
< c(t) + 0(0 + l)(o+l)/Q(O + 1)-(Q+1)/oC(t)

(o+l)c(t)
for
t::::T.
Thus, the sequence {Yn(t)}~=o is monotonically non decreasing and bounded

above by (0+ l)c(t). Hence, {Yn(t)} converges uniformly to a continuous
function y(t), which is a solution of equation (3.9.92).
Lemma 3.9.6. Assume that c(t) and P(s, t) are nonnegative continuous
functions for T S s, t < 00. If there exists E > 0 such that
00
P(s, t)c(o+l)jO(s)ds :::: (0 + 1)-(0+1)/0(1 + E)C(t)
t=
for
t:::: T,
(3.9.93)
then the inequality
y(t) :::: c(t)
+0
IX) P(s, t)ly(s)l(a+1)/ads
does not have a continuous solution y(t).
for
t:::: T
(3.9.94)
195
Proof.
Suppose to the contrary that y(t) is a continuous function
satisfying (3.9.94) for t::>: T. Then, y(t)::>: c(t) ::>: 0 for t::>: T, which
implies y(c<+l)/Ct)::>: C(c<+l)/Ct) ::>: 0 for t::>: T. Thus, we have
y(t)
>
c(t)
>
[1
+a
lOG P(s, t)c(n+1)/0(s)ds
+ a(l + E)(a + 1)-(<>+1)/c<]
c(t)
Continuing this process, we obtain y(t) ::>: !3nc(t),

!3n+1 and
!3n+1 = 1 + a!3~n+1)/<>(a
+ 1)-(0+1)/"(1 + E),
t::>: T.
for
where !31
I, !3n <
n = 1,2,.
(3.9.95)
We claim that
lim
n->oo
!3n =
(3.9.96)
00.
Assume to the contrary that liIll n ->oo!3n = A < 00, where A is a constant.
Clearly, A::>: l. Now, it follows from (3.9.95) that
(3.9.97)
But, it is easy to verify that (3.9.97) is impossible for A::>: l. This contradiction proves (3.9.96). Then, c(t) == 0 for t::>: T, which contradicts
Theorem 3.9.12. If for every t::>: T > to,
c(t) =
and
1=
t
loo q(s)ds
a- 1 / n (s)c(n+1l/n(s)ds ~
> 0
(3.9.98)
1 ) (0+1)/0
00+1
c(t),
then equation (3.l.1) is nonoscillatory.

Proof. By Lemma 3.9.5 the equation
has a continuous solution y(t) on [T,oo). Then, we have
y'(t) = - q(t) - aa- 1/ n (t)ly(t)l(n+1)/c<
for
t::>: T.
Now, from Theorem 3.7.1 equation (3.l.1) is nonoscillatory.
Chapter 3
196
Theorem 3.9.13. Suppose there is a T 2 to such that condition (3.9.98)

holds. If there is an E > 0 such that
t
2 T,
(3.9.99)
Proof. Suppose to the contrary that equation (3.1.1) is nonoscillatory.
Then following the proof of Theorem 3.7.2 there exist a number T1 and
a function 11'(t) E C 1([T1,00),lR) satisfying (3.7.4) on [T],oo). Without
loss of generality let T = T 1 , then it follows from (3.9.98) that for 9 E n,
liminfH= Ag(T, t) =
q(s)ds > -00. Now, from Lemma 3.9.3, we have
a- 1/"(s)lw(s)l(a+l)/ads < 00. Thus, in view of (3.9.98) and (3.9.75),
w(t) satisfies w(t) = c(t) +0' ft= a- 1 /a(s)lw(s)l(a+1)/ads. But, by (3.9.99)
and Lemma 3.9.6, the equation y( t) = c( t) + a ft= a- 1/a (s) Iy( s) I(0+ 1)/ads
does not have a continuous solution, which is a contradiction.
f;
f;
(lx'(t)IO-]x'(t))'
(a: 1)
-(0+1)
(1
+ t)-(<>+I)lx(t)l a - 1x(t)
O.
(3.9.100)
It is easy to check that equation (3.9.100) has a nonoscillatory solution

x(t) = (1 + t)o/(o+l). Since
q(t) = (O':1)a+] (t+1)-(Q+l),
IX) (1
00
q(T)dT) ds =
(1)
a ) (a+1)/a
= (--
(a: 1)
1=
O'-(1+2a)/a(1
+ t)-a
( 1)
(a+1)/a 00
q(s)ds = - q(s)ds.
0'+1
a
t
0'+1
t
This means that the number (1/(0' + l))(<>+I)/a in Theorems 3.9.12 and
3.9.13 is the best possible.
(a+l)/o
3.10. Oscillation CriteriaIntegrable Coefficients

In what follows, we shall a.ssume that
R(t) =
1t
to
a- 1/a(s)ds --+
00
as t --+
00
a.nd Q(t)
00
q(s)ds <
00.
(3.10.1)
197
To prove the main result of this section, we shall need the following lemma.
Lemma 3.10.1. If
JooR'~-1(s)a-1/cs) exp (-(a+1) Q;!"' JS a-l/CO(Q+(~1/C<d~)dS <
00,
(3.10.2)
then
(3.10.3)
for any nonoscillatory solution x(t) of (3.1.1) and for any constant A>
o.
Proof. Let x(t) be a nonoscillatory solution of equa.tion (3.1.1). Without

loss of generality, we assume that x(t) i- 0 for t 2: T 2: to. It follows
from Theorem 3.7.2 that J3.7.4) holds for t 2: T, where wet) is defined
in (3.7.6). Let yet) = It a- 1/<>(s)lw(s)I(<>+1)/"ds for t 2: T. Then for
t 2: T, we have
y'(t) = _a- 1/"(t)lw(t)I(c<+1)/a = -a- 1/a(t)IQ(t)+ay(t)I(Ct+1)/" ::::; O.

Hence, y'(t) + (a+1)(Ct+l)/Ct a-1/a(t)(Q+(t1/Ct y(t) ::::; 0 for t 2: T, which
implies
yet) ::::; yeT) exp
(-(a + 1)(Ct+1)/Ct t a- 1/ Ct (s)(Q+(S1/ Ct dS)
Joo RCt-l(s)a-1/Ct(s)y(s)ds
Joo R Ct (s)a- 1/"(s)lw(s)I(,,+1)/ads < Now, since
<
00,
for t 2: T.
and hence
00.
ln (x(t)
t a-1/Ct(s)lw(s)I(1-Ct)/Ctw(s)ds,
iT
x(T)
by Holder's inequality, we get
1
n
( ~)
x(T)
<
Ra(s)a- 1/Ct(s)lw(s)l(a+1)/ads
(iTt
1/(Ct+1)
a- 1/ Ct (s) ) Ct/("+1)
R(s) ds
,t2:T.
Thus, there exist constants k1 and k2 such that
Ix(t)1 ::::; k2exp(k1(1nR(tCt/(a+l))
for
t2:T.
Therefore, we have
Joo R a- 1(s)a- 1/<>(s)lx(t)I- Ads

2: k2A Joo R<>-1(s)a- 1/<>(s)exp (-Ak 1(1nR(sa/(<>+l)) ds
00.
198
Chapter 3
Theorem 3.10.1. If there is a constant A, 0 < A :::; (0:

that
+ 1)(a+l)/a
jOOR',-1(s)a- 1/a(s) exp ( -A jS a- 1/ CX (E)IQ(E)I(1-a)/aQ(OdE) ds <
such
00,
(3.10.4)
x(t) > 0 for t::;> T ::;> to. From (3.10.4) we find that (3.10.2) holds, and
this in view of Lemma 3.10.1 implies (3.10.3). Since x(t) is nonoscillatory,
it follows from Theorem 3.7.2 that (3.7.4) holds for t::;> T where w(t) is
as in (3.7.6). Hence, for t::;> T,
Proof.
which implies
x(t) ::;> x(T) exp ( [ a-1/<>(s)IQ(s)I(1-</<>Q(S)dS),
t::;> T.
Joo
Thus, from (3.10.4), we obtain

RCt-l(s)a-1/a(s)lx(s)I--'ds <
00,
Joo
Corollary 3.lD.1. If
exp (-2
Q(E)dE) ds < 00, then the linear
equation x//(t) + q(t)x(t) = 0, where q(t) is as in (3.1.1) is oscillatory.
Proof. Take 0:
= 1 and A = 2 in Theorem 3.10.1.
Example 3.10.1. Consider the half-linear equation
where c> (1/(0:
+ 1(<>+l)/a
R(t) = t,
Q(t)
is a constant. Then, we have
_C a ,
0:
(0:
+ 1)(a+1)/a (;-C) 1/a >
0:
and
JOO Ra-l(s) exp ( -(0: + 1)(<>+1)/<> jS Qa(OdE)
ds <
00.
It follows from Theorem 3.10.1 that equation (3.10.5) is oscillatory. In fact,

it has a nonoscillatory solution x(t) = t a/(<>+1) if C = (1/(0: + 1(<>+1)/0.
199

lif~~f R"'(t)Q(t)
>
~a (_a_)Q+1,
0'+1
(3.10.6)

It follows from (3.10.6) that there exist a constant I> (1/0') x
and a T ~ to such that RU(t)Q(t) > I for t ~ T.
Then, we have
Proof.
(0'/(0'
+ 1))"'+1
Thus, by Theorem 3.10.1, equation (3.1.1) is oscillatory.
Remark 3.10.1.
From Example 3.10.1 it is clear that the inequality
(3.10.6) is sharp, i.e., the constant (1/0')(0'/(0'+1))",-1 is the best possible.
Corollary 3.10.3.
such that
If there exist a constant A, 0
lir:linf ta-1/a(t) l'>"IQ(t)IP-n)/aQ(t)
Hoo
1- a
R(t)
< A < (a + 1)(0+1)/",
+ ~a(1-,,)/a(t)a'(t)] >
a
1,
(3.10.7)

Proof. It follows from (3.10.7) that there exist a constant I > 1 and a
T ~ to such that
ta- 1 / n (t) l'>"IQ(t)l(l-n)/nQ(t)
1- a
R(t)
+ 'a(1-n)/a(t)a'(t)l > I
~
for t
Then, we have
Thus, by Theorem 3.10.1, equation (3.1.1) is oscillatory.
Next, we present the following result.

Theorem 3.10.2. Suppose equation (3.1.1) is nonoscillatory and
Then (3.1.1) has a solution x(t) satisfying
J'xo Ix(s)ln+lds =
00.
~ T.
Chapter 3
200

x(t) > 0 for t ~ T ~ to. It follows from Theorem 3.10.1 that (3.10.4)
holds for t ~ T, where w(t) is as in (3.7.6). Then for t ~ T, we have
Proof.
d
-lnx(t)
dt
a-1/O(t)lw(t)I(l-o)/ow(t) ~ a-1/O(t)IQ(t)I(1-O)/OQ(t),
which implies
x(t)
x(T) exp
(t a-1/a(s)IQ(s)I(1-a)/aQ(S)dS) .
Thus, from (3.10.8)' we find that

the proof.
Joo Ix(s)lo+lds =
00.
This completes
Now, we define the sequence of functions {hn(t)}~=o for t ~ to as

follows (if it exists):
1
1 a-l/"'(s)(h~_l(S))("'+l)/"'ds+ho(t),
00
ho(t)
hn(t)
q(s)ds
Q(t)
00
n=1,2,.
(3.10.9)
Clearly, h1(t) ~ ho(t) and this implies that hi(t) ~ ht(t) for t ~ to.
Also, by induction
hn+l (t)
hn(t),
n = 1,2, . . .
for
to,
(3.10.10)
i.e., this sequence {hn(t)} is nondecreasing on [to, 00).

Theorem 3.10.3. If equation (3.1.1) is nonoscillatory, then there exists
a T ~ to such that
lim hn(t)
t ..... oo
= h(t) <
00
for
T.
(3.10.11)
Proof. Suppose equation (3.1.1) is nonoscillatory. Then it follows from

Theorem 3.10.2 that there exists w(t) E Cl([T,oo),1R) such that
w(t)
00
a-1/a(s)lw(s)l(a+l)/ods
00
q(s)ds
for
~ T ~ to.
Thus, w(t) ~ ho(t), and hence w+(t) ~ ht(t) for t ~ T. This implies
that
w(t)
> a
> a
00
1
1
00
00
a-1/"'(s)lw(s)I(O+l)/ads +
00
q(s)ds
a-1/"'(s)[w+(s)](,,+1)/"'ds + ho(t)
a-1/"'(s)[ht(s)] (<>+l)/"'ds
+ ho(t) = h1(t)
for t
~T
201
and, by induction
w(t)
hn(t),
= 0, 1,""
T.
(3.10.12)
Now, from (3.10.10) and (3.10.12), we find that the sequence {hn(t)} is
bounded above on [T,oo). Hence condition (3.10.11) holds.
Corollary 3.10.4. Equation (3.1.1) is oscillatory, if either

(i)
there exists a positive integer
such that
n = 1,2"", m - 1, but hm(t) does not exist, or
hn(t) is defined for
(ii) hn(t) is defined for n = 1,2""

but for arbitrarily large T* ~ to
there is a t* ~ T* such that limn-too hn(t*) = 00.
Next, we define the sequence of functions {gn(t)}~o for t ~ to as
follows (if it exists):
1 =
0' 1 a~l/a(s)[g~_l
00
go(t)
ql(s)ds
Ql(t)
00
gn(t)
(s)](a+l)/ads
+ Igo(t)l,
n = 1,2,.
(3.10.13)
Clearly, 91(t) ~ 190(t)1 ~ g(j(t) for t ~ to. Also, by induction
gn+1(t) 2 gn(t),
n = 1,2 ..
and
t ~ to,
(3.10.14)
i.e., this sequence {gn (t)} is nondecreasing on [to, (0).

In what follows, we shall assume that
Rl(t) =
t a~l/a(s)ds
ltD
Theorem 3.10.4.
-+
00
as t -+
00
and Ql(t)
<
00.
(3.10.15)
Equation (3.2.15) is nonoscillatory if there exists a
T 2: to such that
lim gn(t) = g(t)
n-too
<
00
for
t ~ T.
(3.10.16)
Proof. If (3.10.16) holds, then from (3.10.14) and (3.10.16) it follows that
gn(t) ::; g(t), n = 0,1, . .. for t ~ T. Applying the monotone convergence
theorem, we have
g(t)
0'
1 a~l/a(s)[g+(s)](a+l)/Qds +
00
Thus, we get
g+(t)
1 a~l/a(s)[g+(s)]("'+1)/Qds +
10' 1 a~l/Q(s)[g+(s)](a+l)/ads +
0'
= g(t)
>
Igo(t)1 2: O.
00
00
Igo(t)1
go(t)l
Chapter 3
202
Clearly, g+(t) E C([T, CXJ), m+), and now it follows from Theorem 3.10.2
that equation (3.2.15) is nonoscillatory. This completes the proof.
Corollary 3.10.5. If equation (3.2.15) is oscillatory, then either
(i)
there exists a positive integer m
such that
gn (t) is defined for
= 1,2" ", m - 1, but gm(t) does not exist, or

(ii) gn(t) is defined for n = 1,2"", but for arbitrarily large T* 2: to,
there exists t* 2: T* such that lim n --+ oo g" (t*) = CXJ.
Now, if a(t) = al(t) and ho(t) = go(t) = Q(t) = fl oo q(s)ds 2: 0 for

t 2: to, then hn(t) = gn(t) 2: 0, n = 0,1,2,. Thus, the following two
corollaries hold:
Corollary 3.10.6.
Suppose Q(t) 2: 0 for t 2: to. Then equation
(3.1.1) is nonoscillatory if and only if there exists T 2: to such that
lim n --+ CXl h n (t) = h( t) < CXJ for t 2: T.
Corollary 3.10.7. Suppose Q(t) 2: 0 for t 2: to. Then equation (3.1.1)
is oscillatory if and only if either
(i) there exists a positive integer m such that

n = 1,2" . " m - 1, but hrn(t) does not exist, or
hn(t) is defined for
(ii) hn(t) is defined for n = 1,2" . " but for arbitrarily large T* 2: to
there is a t* 2: T* such that limn--+= hn(t*) = 00.
Now, using Theorems 3.10.3 and 3.10.4, we shall prove the following
result.
Theorem 3.10.5.
o<
Assume that
al(t) .:::: a(t),
IQ(t)1 .:::: Ql(t) for all sufficiently large t. (3.10.17)
If (3.2.15) is nonoscillatory, then equation (3.1.1) is nonoscillatory, or equivalently, if (3.1.1) is oscillatory, then equation (3.2.15) is oscillatory.
Proof. Suppose equation (3.2.15) is nonoscillatory. It follows from Theorem 3.10.3 that there exists a T 2: to such that
lim gn(t) = g(t) <
n--+oo
00
for
t 2: T.
(3.10.18)
Clearly, by (3.10.17), ho(t)':::: Iho(t)1 = IQ(t)1 .:::: Ql(t) = go(t),

hri(t) .:::: gri(t) for t 2: T. Thus, for t 2: T,
hl(t)
Cl'
<
Cl'
1= a~l/a(s)[hri(s)J(O+1)/Ods
1= a~l/Q(s)[gt(s)J(Q+l)/Qds
and hence
+ Iho(t)1
+ go(t)
gl(t).
203
Also, by induction
hn(t) ::; gn(t),
n = 0, 1,2,
and
(3.10.19)
T.
Therefore, by (3.10.14), (3.10.18) and (3.10.19), we have
h(t) =
lim hn(t)::;
n~CX)
lim gn(t)
n-too
= g(t) <
00
for
~ T.
Hence, by Theorem 3.10.4, equation (3.1.1) is nOlloscillatory. This completes the proof.
Next, we present the following criterion for the oscillation of equation

(3.1.1) when condition (3.7.16) holds.
Theorem 3.10.6. Let A > a be a constant. If there exists a number c
such that
li~~p 7T<>-A(t)
00
7T A(S)q(s)ds > c > A _ a
A )<>+1
+1
' (3.10.20)

Proof. Assume to the contrary that equation (3.1.1) is nonoscillatory. We
define the function w(t) as in Theorem 3.7.4. Then there exists a T ~ to
such that (3.7.26) holds for t ~ T. In view of the boundedness of 7TO(t)W(t)
(cf. Lemma 3.7.3), we find that 7TA(~)W(~) = 7TAO(~)7TO(~)W(~) ~ 0 as
~ ~ 00, and
and
1[00 a- 1!0(s )7T A(s) Iw(s) 1(<>+l)!<>dsl
: ; [= a- 1!0(s)7T A- O- 117T<>(S)w(s)I(0+1)!ods <
for all t ~ T. Therefore, letting ~ ~
00
00,
in (3.7.26), we find
[= 7T A(S)q(s)ds + 1= a- 1!0(S)7T A- 1(S)w(s)ds

+a
00
I7TO(S)w(s)I(<>+!)!o
a- 1!<>(S)7T A(s)lw(s)I(<>+1)!Ods
+ -7T<>(S)W(S) + _1
a
(_A
a+1
for
)0+1
~ T.
> 0,
Chapter 3
204
which implies
00
7r A(s)q(s)ds - - - - A-a a+1
),,+1 7r -"(t).
A
Hence, we get
Now, it follows from (3.10.20) that limsuPHoo 7r"(t)w(t) > 0,

contradicts (3.7.19). This completes the proof.
which
Corollary 3.10.8. Let A > a be a constant. If

Theorem 3.10.7. If
limsup
t-+oo
1t
[7r"(S)q(S) _
to
1/"(S)7r-1(S)] ds = 00,
(_a_)"+l
+
a
a-
(3.10.21)
Proof. Assume to the contrary that equation (3.1.1) is nonoscillatory. We
define the function w(t) as in Theorem 3.7.4. Then there exists a T ~ to
such that (3.7.26) holds for t ~ T with A = a. By Lemma 3.7.2, we have
~
0,
which implies
-00,
Now, it follows from (3.10.21) that liminfHoo 7r"(t)w(t) =

which
contradicts the fact that 7r"(t)w(t) ~ -Ion [T,oo). Thus, equation
Theorem 3.10.8. Let A < a be a constant. If
limsup 7r"-A(t)1 7r A(s)q(s)ds > 1+

t
t-+oo
to
a~A (a+A 1)"+1,
(3.10.22)
205
Proof. Suppose equation (3.l.1) is nonoscillatory. As in Theorem 3.10.6,

(3.7.26) holds for t::::: T for some T 2: to. By Lemma 3.7.2, we have
A
In"(s)w(s)l(a+1)/a + -nG(s)'w(S)
+ -1 ( -A- )
a
0+1
a+ 1
2: 0,
which implies
Then, we find
Hence, we obtain
limsup n"-"(O
[,-+1'
l'
nA(s)q(s)ds < 1 + - -
a - A
-A- )0+1
a +1
'
which contradicts (3.10.22). Thus, equation (3.l.1) is oscillatory.
3.11. Oscillation of Damped

and Forced Equations
In this section, we shall discuss the oscillatory behavior of the damped
equation
(lx'(t)l a- 1x'(t))' + q(t)F(x(t), x'(t)) = 0
(3.11.1)
and the forced equation
(lx'(t)la-1 X '(t))' + q(t)lx(t)la- 1x(t)

where
(i)
a > 0 is a constant,
(ii) e(t), q(t) E C([to, 00), JR), and
(iii) FE C(JR 2 , JR) and sgn F(x, y)
= sgn x.
e(t),
(3.1l.2)
Chapter 3
206

Theorem 3.11.1.
Suppose the function F satisfies the condition
(-1),,+1 = 1, F(AX,AY) = N"F(x,y) for all (A,X,y) E lR?, and
dw
a F(l, w)
If
00
> c for every v
q(s)ds =
to
00,
E IR.
(3.11.3)
~0
(3.11.4)
to

Proof. Suppose x(t) # 0 is a nonoscillatory solution of equation (3.11.1),
then since - F(-x, -y) = F(x,y), we can assume that x(t) > 0 for
t ~ to ~ O. Define
Ix'(t)I"-l X'(t)
x"(t)
w(t)
for
t ~ to.
Then for t ~ to, we have w'(t) +alw(t)I("+l)!" +q(t)x-"(t)F(x(t), x'(t

= 0, or w'(t) + odw(t)l(o+l)}" + q(t)F(l, w(t = 0, and hence
w'(t) : : : - q(t)F(1,w(t
for
t ~ to.
(3.11.5)
Dividing both sides of (3.11.5) by F(l,w(t > 0, we get
(t)
w(to)
- v() : : : _
it
F 1, v
to
q(s)ds -+ _
00
as
t -+
00,
(3.11.6)
Remark 3.11.1. When condition (3.11.3) does not hold the argument
presented above shows that every solution of (3.11.1) is either oscillatory
or tends to zero monotonically as t -+ 00. In fact, in this case inequality
(3.11.6) yields
lim w(t) =
t-+oo
lim
t-+oo
1x' (t) 1,,-1 x' (t)

x"(t)
00,
so there exists a constant k > 0 such that x'(t)/x(t):::::: -k for all t

sufficiently large, say, t ~ T ~ to which on integration gives 0 < x(t) <
x(T) exp( -(t - T for all t ~ T.
Next, we shall prove the following result.
Theorem 3.11.2. Suppose q(t) ~ 0 eventually, and
F(AX,AY) = Af3 F(x,y)
and
(_1)f3+ 1 = 1
(3.11.7)

for all ()..,x,y) E 1R3 , where j3 >
L= (1=
207
is a constant. If
q(u)du r/<> ds
00,
(3.11.8)

x(t) > 0 for t ~ to ~ O. Since (lx'(t)I<>-l x'(t))':::; 0 for t ~ to, it is
easy to check that x'(t) > 0 for t ~ t1 ~ to, and
Proof.
lim x'(t) =
Hoo
x(t)
o.
(3.11.9)
and consequently, from the continuity of F(x, y), given a fixed positive
constant E, E < F(l, 0), there exists a t2 ~ t1 such that
k = F(l,O) for every t
x'(t)) :::; F(l,O)

< F ( 1, x(t)
+E
(3.11.10)
to. Now, we have
x'(t)
(1= q(s)F(x(s), x'(S))dS) 1/<>

(1 q(s)x~(s)F (1, ~(~}) dS) 1/<>
(F(l,O) _
(1= q(S)X~(S)dS)
00
>
or
x'(t)
xf3iot)
E)I/o<
k1/0< (
I/o< ,
(t= q(S)dS) 1/0
it
(3.11.11)
Integrating (3.11.11) from t2 to t, we find
k 1/<> i t ( (= q(U)dU) 1/<> ds :::;

t2
is
(t)
v-~/O<dv
<
00,
X(t2)

Theorem 3.11.3. Let q(t)
liminf
t-->=
{t e(s)ds = _ 00,
iT
0 for t ~ to. If for every T ~ to,

lim sup
t-->=
{Tt e( s )ds =
i7
00,
(3.11.12)
208
Chapter 3
and
liminf
t~oo
lim sup
t~oo
1
1
t
a-1/Ot(s)
(1S e(u)du)
(1S e(u)du)
l/Ot ds
a-1/Q(s)
00,
(3.11.13)
l/Ot ds
00,

Proof. Assume the contrary. Then without loss of generality, we can
assume that there is a nonoscillatory solution x(t) of (3.11.2), say, x(t) >
o for t::::: t1 ::::: to. From (3.11.2), we have (a(t)lx'(t)IOt-1x'(t))' e(t)
for t::::: t1. Thus, it follows that
s:
a(t)lx'(t)IOt-1X'(t) - a(t1)lx'(t1)l',,-lx'(t1)
s:
it
e(s)ds.
(3.11.14)
t,
By (3.11.12) there exists aT::::: to sufficiently large so that x'(T)

and x'(t) < 0 for t::::: T. Replacing h by T in (3.11.14), we get
x'(t)
s:
a-1/Ot(t)
and
x(t)
s:
x(T)
+[
Therefore, liminft~oo x(t)

o eventually.
Theorem 3.11.4.
82 < t2 such that
a- 1/ Ot (8)
-00,
e(s)ds
<0
)l/Ot
(lS e(U)du)
l/Ot ds.
which contradicts the fact that x(t) >
Suppose for every T::::: 0 there exist T
s: Sl < t1 s:
(3.11.15)
Denote by
If there exists u E D(Si' ti) such that
(3.11.16)
for i
= 1,2
and (_l) Ot +l
I, then equation (3.11.2) is oscillatory.
209
Proof. Suppose x(t) is a nonoscillatory solution of equation (3.11.2), say,

x(t) > 0 for t::::: to for some to depending on the solution x(t). Define
w(t) = -a(t)lx'(t)I<>-lx'(t)/x O(t) for t::::: to. It follows from (3.11.2) that
w(t) satisfies the first order nonlinear (generalized Riccati) equation
w'(t) = aa- l / O(t)lw(t)l(a+l l /<>
+ q(t) -
e(t)
xo(t)
t::::: to.
for
(3.11.17)
By assumption, we can choose 81, tl ::::: to so that e(t) ::; 0 on the interval
SI < tl. On the interval I, w(t) in view of (3.11.15)
satisfies the differential inequality
1= [SI' hJ where
w'(t) ::::: aa-1/<>(t)lw(t)l(a+l l /<> + q(t)
on
I.
(3.11.18)
Let u(t) E D(SI, td be as in the hypothesis. Multiplying (3.11.18) by

u<>+l(t) and integrating over I, wefind
j u<>+I(s)w'(s)ds ::::: j aa- l /O(s)lw(s)I(<>+I)/<>ds + j uO+l(s)q(S)d8.
(3.11.19)
Integrating (3.11.19) by parts and using the fact that U(Sl) = u(h) = 0,
we obtain
-(a+1)
j uO(s)u'(s)w(s)ds ::::: j aa- l /<>(s)lw(s)1 0;;' ds+ j u<>+I(s)q(s)ds,
or
>
j u<>+I(s)q(s)ds + j [aa- l /<>(s)uO+l (s)lw(s)I(<>+l)/o

-(a
+ l)uO(s)lu'(s)llw(s)l]
ds.
Let AA = aa- 1/<>(s)u<>+l(s)lw(s)l(o+ll/<>, >. = (a + l)/a and B =

((aa(sW/(o+l)lu'(s)I)", then from Lemma 3.2.1, we obtain for t1 > s:::::
Sl ::::: to,
aa- 1/<>(s)u o+l(s)lw(s)l(o+l)/<>

- (a
+~
((aa(s))I/(O+llu'(s)r+ 1
+ l)u<>(s)lu'(s)llw(s)1
Thus, we have 0::::: Q1(t)
:::::
o.
> 0, which contradicts (3.11.16).
When x(t) is eventually negative, we use u E D(S2, t2) and e(t) ::::: 0
on [S2, t2J to reach a similar contradiction. This completes the proof.
Remark 3.11.2. If a
can be replaced by
Qi(U)
l~i
1 in Theorem 3.11.4, then condition (3.11.16)
[q(s)u 2(s) - a(s)(u'(s)?J ds ::::: 0,
1,2.
Chapter 3
210
Example 3.11.1. Consider the forced half-linear equation
(3.11.20)
Here, the zeros of the forcing term sin0 are (mT)2. Let u(t) = sin 0.
For any T 2': 0 choose n sufficiently large so that (mT? 2': T, and set
81 = (mT)2 and tl = (n + 1)21T2 in (3.11.16). It is easy to verify that
(n+l)271'2 [
(n7l')2
1
]
sin4 .jS - ---;=;. cos 4 .jS d8
2y 8
168y 8
r;.
l~nH)7I' [sin4'V - 8~12 cos4 V] dv

>
r(nH)7I'
in
[~
4
71'
~1T _
4
cos 2v _ cos 4v __1_] dv

2
2
8v 2
8n(n + 1)1T
> 0
Similarly, with 82 = (n + 1)21T2 and t2 = (n + 2)21T2 we can show

that Q2(U) > O. It follows from Theorem 3.11.4 that equation (3.11.20) is
oscillatory.
3.12. Distance of Zeros of Oscillatory Solutions

Here, we shall derive lower bounds for the distance between consecutive
zeros of solutions of the half-linear differential equation
(lx'(t)la:-lX'(t))'
+ q(t)lx(t)la:-lX(t) =
0,
>0
(3.12.1)
where q: [to, (0) ---+ lR is locally integrable for some to 2': O.

To establish main results, we shall need the following lemmas.
Lemma 3.12.1. Suppose PI, P2 E C([a,b],lR) satisfy Pl(t) :::;P2(t) for
all t E [a, b]. Let (PI and <P2 defined on [a, b] be respective solutions of
the integral equations
(3.12.2)
and
<P2(t)
= P2(t) + k
it
1<P2(tW'd8,
(3.12.3)
where k > 0 and A> 1 are constants. If <Pl(t) 2': 0 on [a,b],

<Pl(t) :::; <P2(t) for all t E [a,b].
then
211
Proof. First, consider the case PI (t) < P2(t) for all t E [a, b]. We
shall show that cPl(t) < cP2(t) on [a, b]. Suppose to the contrary that
cPl(t*) = cP2(t*) for some t* E [a, b]. Since cP2(a) = p2(a) > Pl(a) = cPl(a),
there must be a smallest t, say, tl > a, tl :'::: t* such that cP2 (tl) = 4>1 (tl)
and <!>2(t) > cPl(t) on [a, td.
Let G(y) = Iyl\ then G(y) is nondecreasing with respect to y 2:

It follows that
cPl(t 1 )
Pl(t 1 )
+k
< P2(t 1 ) + k
ltl
ltl
o.
G(cPl(s))ds
G(cP2(s))ds = cP2(tt},
which contradicts the fact that <!>l(td = <!>2(td. Hence, <!>l(t) < cP2(t) for
all t E [a, b]. Now, a continuity argument completes the proof.
Lemma 3.12.2. If x(t) is a solution of (3.12.1) satisfying x'(c) = 0, x(b)

= 0, x(t) > 0 and x'(t):,::: 0 for t E (c, b), then sUPc<t O.
Proof.
Q(t)

q(s)ds, and define
J:
J: q(s)ds
:'::: 0 for t
[c, b]. Let
(3.12.4)
Then, we find
q(t)
w'(t)
which implies
w(t)
Q(t)
+a
+ alw(t)l(o+l)/a, t E [c,b)
it
Iw(s)l(a+l)/ads,
(3.12.5)
[c, b)
(3.12.6)
and clearly, we have

w(c) = 0,
lim w(t) =
t ..... b-
00
and
w(t) 2: 0,
(c, b).
(3.12.7)
Similarly, if z(t) is a nontrivial solution of the equation (lz'(t)la-l z'(t)), =

with z'(c) = 0, we let Wl(t) = -Iz'(t)l<>-lz'(t)/ (lz(t)la-lz(t)), then
Wl(t) = 0 for all t E [c,b]. Since
Wl(t)
it
IWl(S)I(a+l)/ads,
t E [c,b)
it follows from Lemma 3.12.1 that w(t):,::: Wl(t) = 0 for all t E [c,b).
This contradicts the fact that limt ..... b- w(t) = 00.
Chapter 3
212
Theorem 3.12.1. Let x(t) be a nontrivial solution of equation (3.12.1)

satisfying x'(e) = 0, x(b) = 0 and x(t) # 0 for all t E [e, b). Then,
(b - e)a sup
c:::;t~b
It
. c
q(s)dsl > l.
(3.12.8)
Moreover, if there are no extreme values of x(t) in (e,b), then
(b - e)a sup
c~t~b
It
q(s)ds > l.
(3.12.9)
. c
Proof. Without loss of generality, we assume that x(t) > 0 for all
t E [e, b). Define w(t) as in (3.12.4), and let
cjJ(t)
it
Iw(s)l(a+l)/ads
for
t E [e, b).
(3.12.10)
for
t E [e,b),
(3.12.11)
w(t) = cjJ(t)
where Q(t) =
limHb- cjJ(t) =
I: q(s)ds.
Thus, w(c) = cjJ(c) = 0 and
limHb- w(t) =
00.
Now set Q*
cjJ'(t)
+ Q(t)
= sUPc~t9IQ(t)l,
so that
alw(t)l(a+l)/a < a(Q*
+ cjJ(t))(a+l)/a
for
t E [e, b)
and
a(Q*
cjJ'(t)
+ cjJ(t))(a+l)/a
::; 1 for
tE[c,b).
(3.12.12)
Integrating the inequality (3.12.12) from e to b and using limHb- cjJ(t) =

00, we obtain
(Q*+~(t))l/a[:
::; b-e,
which implies that

(3.12.13)
We remark that the equality in (3.12.13) cannot hold, for otherwise, IQ(t)1
II: q(s)dsl
Q* for t E [e, b), which contradicts the fact that Q(t) is

continuous and Q(e) = O. Thus, inequality (3.12.8) holds.
=
I:
If e is the largest extreme point of x(t) on [e, b), then x'(t)::; 0

and thus w(t) ~ 0 for t E [e, b). Set Q* = sUPc:::;t 0 and x'(t)::; 0 on (e,5), it follows from
Oscillation and nonoscillation of baH-linear differential equations
213
Lemma 3.12.2 that Q* > O. Hence, by (3.12.11), 0::::; w(t) ::::; Q* + 4>(t).
The rest of the proof is similar to that presented above, and hence we omit
the details.
In a similar way, we can prove the following result.

Theorem 3.12.2. Let x(t) be a nontrivial solution of equation (3.12.1)
satisfying x(a) = 0, X'(T) = 0 and x(t) =I- 0 for all t E (a, T]. Then,
I.Itr
(T - a),' sup
a::;t::;T
q(s)dsl > l.
(3.12.14)
Moreover, if there are no extreme values of x(t) in (a, T), then
iT
(T - a)" sup
a::;t::;T
q(s)ds > l.
(3.12.15)
Definition 3.12.1. We say that (3.12.1) is right disfocal (left disfocal) on

[a,b] if the solutions of equation (3.12.1) such that x'(a) = 0 (x' (b) = 0,
respectively) have no zeros in [a, b].
(b - c)'" sup
a::;t::;b
lib
c
q(S)dSI ::::; 1,
then equation (3.12.1) is right disfocal on [c,d), if

(c - a)'" sup
a::;t::;c
II
a
q(s)dSI ::::; 1,
then equation (3.12.1) is left disfocal on (a, c].

In what follows for convenience, we shall let TJ
= min{ 4, 4"'}.
Theorem 3.12.3. Let a and b, a < b denote consecutive zeros of a

nontrivial solution x(t) of equation (3.12.1). Then there exist two disjoint
subintervals, say, hand h of [a, b] satisfying
(b-a)"j
[,U!2
and
J[a,b]\(Il uJ,)
Proof. Let c and
q(s)ds > TJ
q(s)ds ::::; O.
(3.12.16)
(3.12.17)
denote the smallest and largest extreme points of
x(t) on [a, b], respectively. (If there is only one zero of x'(t) in (a, b),
Chapter 3
214
then c and T coincide). Thus, x'(c) = 0, x(b) = 0 and x'(t) =f. 0 for all
t E (c, b]. By Theorem 3.12.1 inequality (3.12.9) holds. Thus, there exists
bI E (c, b] such that
(b - c)<>
b,
q(s)ds > 1
f b, q(s)ds
and
fb
(3.12.18)
:2: . c q(s)ds.
Similarly, it follows from Theorem 3.12.2 that there exists al E [a, T) such
that
(T-a)o:lT q(s)ds > 1
11' q(s)ds
and
a,
Let h = [c, bI ] and h = [aI, T].

cases:
Case 1. If
0:
11' q(s)ds.
:2:
a,
Now, we consider the following two
> 1, then
(b-a)o:j
I , Uh
q(s)ds :2: [(b - c)
+ (T - a)]'"
(fbI
c
q(s)ds +
11'a, q(S)dS)
> [(b - c)" + (T - a)"] [(b ~ c)<> + (T

:2:
Case 2. If 0 <
(b-a) ( j
(3.12.19)
I , UI 2
0:
~ a)" ]
1
0/2
1
]
~(b-c) 0/2 (b-C)<>/2
+ (T-a) (T-a)<>/2
- 4.
< 1, then
q(S)dS)
I/o
= (b - a)
l'
b
q(s)ds +
?: (b-a)
b
)
[ (l'q(s)ds
?: [(b-c)+(T-a)]
11'a, q(s)ds
1/<>
1/"
U~q(S)dS)
[(b~C) + (T~a)]
It follows from the cases 1 and 2 that (b - a)<>
J:,
JI,UI 2
q(s)ds >
J:'
'T].
1/<>]
?:4.
Next,
by (3.12.18) and (3.12.19), we have

q(s)ds ~ 0 and
q(s)ds ~ O.
Thus, to verify (3.12.17) it suffices to show that
q(s)ds ~ O. In fact,
since X'(T) = x'(c) = 0, W(T) = w(c) = 0 from (3.12.6), we find
o=
w(c) - W(T) =
This means that
J: q(s)ds
q(s)ds + 0:
J:
Iw(s)l(a+l)/ods.
~ 0, which implies that (3.12.17) holds.

Corollary 3.12.2.
h of [a,b],
215
Suppose for every two disjoint subintervals hand
(b - a)"
, I1UI,
q(s)ds:S
(3.12.20)
'I].
Then each solution of equation (3.12.1) has at most one zero on [a, b].
Proof. Suppose to the contrary, there exists a nontrivial solution x(t) of
equation (3.12.1) with X(7) = x(c) = 0 for a:S 7 < c :S b. Without loss
of generality, we assume that x'(t) i- 0 for t E (7, c). By Theorem 3.12.3
there exist two disjoint subintervals hand 12 on [7, c) C [a, b] such
that (C-7)a hUI, q(s)ds > '1]. Thus, we have (b-a)" fhuI, q(s)ds > '1],
Corollary 3.12.3. Suppose a nontrivial solution x(t) of equation (3.12.1)

has N::::: 2 zeros in [a, b]. Then there exist 2(N -1) disjoint subintervals
of [a, b], Iij where i = 1,2"", N - 1, j = 1,2 such that
N <
[(b-'I]a)c< jq(S)dsf/ +1
(3.12.21)
(3.12.22)
and
[a,b]\!
q(s)ds < 0,
2
I
h
I = UiN-l
were
=1 Uj =1 ij.
Proof. Let ti, i = 1,2, .. , N be the zeros of x(t) in [a, b]. By Theorem
3.12.4 for each i = 1,2, ... , N - 1 there are two disjoint subintervals, say,
IiI and Ii2 of [ti, ti+l] such that
i I il UI i2
and
q(s)ds >
(3.12.23)
'I]
(ti+l - ti)"
q(s)ds :S O.
(3.12.24)
i[ti,ti+d\(Iil UIi,)
Summing (3.12.23) for i from 1 to N - 1, we get
iI q(s)ds >
N-l
'I]
> 'I](N -
1
(ti+l - ti)a
1]
1
[
1)
iI)<> ...
1]
[ 1
(t2 -
1/(N-l)
(tN - tN_l)a
l/a(N-l)
'I](N - 1) (t2 - tl) ... (tN - tN-d
>
'I]
(N
1)<>+1
(tN - tl)et'
Chapter 3
216
which implies (N - 1),,+1 < ((b - a)")/rJ II q(s)ds. This implies that
(3.12.21) holds. From (3.12.24) it is easy to deduce (3.12.22). This com
pletes the proof.
(Ix' (t) I,,-lx' (t)'
+ (sin kt) Ix(t) I"-lx(t) =
0,
(3.12.25)
where k > 0 is a constant. Then,

(i) equation (3.12.25) is right disfoca.l on [0, b) and left disfocal on (0, bj
if 0 < b < k/2, and
(ii) each solution of (3.12.25) has at most one zero in [0, bj if 0 < b <
(kTJ/4)1/" .
(i)
Since q(t)
sup
099
= sinkt and 0 < b < k/2,
I (t q(S)dSI
io
sup
099
II;
It
io
sinksdsl <
k'
Thus, bSUPO:St:Sb
q(s)dsl :::; 1. Now, by Corollary 3.12.1 equation
(3.12.25) is right disfocal on [0, b). Similarly, (3.12.25) is also left disfocal
on (0, bj.
(ii) It follows from 0 < b :::; (k'17/4)1/" that for any two disjoint
subintervals hand 12 of [0, bj, II , uI2 q(s)ds = II , uI2 sinksds :::; 4/k,
which implies b" f h uI2 q(s)ds :::; TJ. Now, by Corollary 3.12.2 each solution
of equation (3.12.25) has at most one zero on [0, bj.
In what follows for any t and 0 > 0, we shall denote by (h u12 )(t, 0)
the union of two disjoint subintervals hand h of [t, t + oj.
Theorem 3.12.4. Let x(t) be an oscillatory solution of equation (3.12.1).
If
lim sup
t~oo
(0" 1
(l,UI2 )(t,8)
q( s )dS)
<
'17
(3.12.26)
for all 0 > 0 and for every two disjoint subintervals hand h of
[t, t+oj, then the distance between consecutive zeros of x(t) is unbounded
as t -+ 00.
Proof. Suppose to the contrary that equation (3.12.1) has an oscillatory
solution x(t) whose zeros {tn}~=l contain a subsequence {tnk}~l such
that 0 < t nk + 1 - tnk < 0 for some 0 > 0 and all k. Then by Theorem
3.12.3 there are two disjoint subintervals h (tnk' 0) and h(tnk' 0) of
[t nk , t nk +,] satisfying
217
This implies that
bo.l
q(s)ds > TJ
k,
for all
(h U!Z)(t"k ,6)
Theorem 3.12.5. Let x(t) be an oscillatory solution of equation (3.12.1).

If there exists a 00 >
such that
lim
t-+oo
(IluI~)(t,.50)
q(s)ds =
(3.12.27)
for every two disjoint subintervals hand h of [t, t + bo], then the
distance between consecutive zeros of x(t) is unbounded as t --+ 00.
and any two disjoint
q(s)ds = 0.
(3.12.28)
Proof.
We will first show that for all
subintervals hand h of [t, t + bo],
lim
t-+oo
(hU!z)(t,.5)
0 >
Let k denote the least integer with Ho;::: 0 and t; = t

0,1,"" k - 1 and tk = t + 0, then
L1
+ ioo,
i =
k-l
(IIUlz)(t,.5)
q(s)ds =
i=O
(IiI U1.,)(ti,.50 )
q(s)ds,
(3.12.29)
where l;j(t;,oo)=lj(t,o)n[t;,t;+l] fori=0,1, .. ,k-1 and j=I,2.

Noting that t; --+ 00 as t --+ 00, it follows from (3.12.27) that
lim
t-+oo
q( s )ds = 0,
i = 0, 1, ... , k - 1.
(3.12.30)
(IiI U!i2)(t i ,<lo)
Combining (3.12.29) and (3.12.30), we obtain (3.12.28) which implies

lim
t-+oo
(00.1
(II UI2)(t,.5)
q(S)dS)
for all b > 0 and any two disjoint subintervals hand 12 of [t, t
The result now follows from Theorem 3.12.4.
+ 0].
Remark 3.12.1. Consider the general half-linear equation (3.1.1). By

direct computation it is easy to verify that the change of variables (t, x) --+
(T,X) given by T = T(t) =
a-1/Ds)ds for t;::: to and X(T) = x(t)
transform (3.1.1) into the equation
It:
(3.12.31)
Chapter 3
218
where Q(7) = a1/0t)q(t), (- = djd7) , t = t(7) is the inverse function

of 7 = 7(t) given above. Since equation (3.12.31) is of the form (3.12.1)
it follows from Theorems 3.12.4 and 3.12.5 that the following oscillatory
criteria for equation (3.1.1) hold.
Theorem 3.12.6.
(3.1.1). If
Let x(t) denote an oscillatory solution of equation
limsup (<>e>
t---+oo
for all <>
(I, UI2)(t,li)
q(S)dS) <
(3.12.32)
1]
> 0 and for every two disjoint subintervals hand I2 of
[t, t+<>], then the distance between consecutive zeros of x(t) is unbounded
as t -+ 00.
Theorem 3.12.7. Let x(t) denote an oscillatory solution of equation

(3.1.1). If there exists a <>0 > 0 such that
lim
t---+oo
(I,Uh)(t,li o )
q(s)ds = 0
(3.12.33)
for every two disjoint subintervals hand h of [t, t + <>0], then the
distance between consecutive zeros of x(t) is unbounded as t -+ 00.
3.13. Oscillation and Nonoscillation

of Half-Linear Equations with
Deviating Arguments
This section is devoted to the study of the oscillatory behavior of halflinear functional differential equations of the type
(a(t)lx'(t)lo<-lX'(t))' + q(t)lx[g(t)]I i3 - 1x[g(t)] = 0
(3.13.1)
(a(t)lx'(t)lo<-lX'(t))' - q(t)lx[g(t)]lo<-lX[g(t)]
(3.13.2)
and
0,
where
(i)
(ii)
IX
> 0, (3 > 0 are positive constants,
a(t) E C([to, (0), JR+) and
Joo
a-1/0s)ds
00,
(iii) q(t) E C([to, (0), JRo), q(t) =t 0 for all large t,

(iv) g(t) EC1([to,00),JR), g'(t);:::O for t;:::to and limHoog(t) =00.
The half-linear ordinary differential equation
(lx'(t)lo<-lX'(t))'
q(t)lx(t)IO<-lX(t)
(3.13.3)
219
Oscillation and non oscillation of haH-linear differential equations
which is the same as (3.13.2) with a(t) == 1 and g(t) == t is nonoscillatory

in the sense that all of its solutions are nonoscillatory, see Elbert [15].
However, the presence of a deviating argument g(t) =f= t in (3.13.2) may
generate oscillation to some or all of its solutions. We illustrate this in the
following example.
Example 3.13.1. Let Sa(t) denote the solution of the equation
(lx'(t)1 0
1 x'(t))'
+ alx(t)la- 1 x(t) =
0,
a> 0
satisfying the initial conditions x(O) = 0, x' (0) = 1. Elbert [15] has shown
that Sa(t) exists uniquely on 1R and it is periodic of period ?T a ,
2';1/ sin (a~l)'
Furthermore, Sa(t) satisfies Sa(t-?To) =

JR.. It follows that So(t) is an oscillatory
solution of both the functional differential equations
where ?Ta =
S",(t + ?T",) = -S",(t) for t
and
3.13.1. Some Useful Lemmas
J:
In what follows, we shall let R[t, s] =

a- 1 /a(u)du for t 2: s 2: t
and R(t) = R[t, to]. To prove our main results we shall need the following
lemmas.
Lemma 3.13.1.
Assume that a, 9 E C([to,oo),1R+), g(t) < t for
t 2: to and limHoog(t) = 00. Let x E C 1 ([to,00),JR.+) be such that
a1/a(t)x'(t) E C 1 ([to,00),1R+) and (a(t)(x'(t))a)':s: 0 for t 2: T 2: to.
Then,
(I)
for each constant k, 0 < k < 1 there is a Tk 2: T such that either
x[g(t)] 2: k
(:(~)) l/a (g~t)) x(t)
for
a'(t):s: 0,
t 2: Tk 2: T,
(3.13.4)
or
x[g(t)] 2: k
(g~t)) x(t)
for
a'(t) 2: 0,
t 2: Tk 2: T,
(3.13.5)
t 2: T.
(3.13.6)
(II) for all sufficiently large T 2: to,
x(t) 2: a1 /a(t)R[t, T]x'(t),
220
Chapter 3
(I) It suffices to consider only those t for which get) < t. Since
al/e>(t)x'(t) is decreasing, it follows from the mean-value theorem that for
each t> get) ~ T, there exists a tl E (g(t), t) such that
Proof.
x(t) - x[g(t)]
a-l/e>(tl) (al/e>(tl)x'(t l )) [t - get)]
:s
a-l/e>(tl)al/e> [g(t)]x'[g(t)][t - get)].
Since x(t) > 0, we get
:s 1 + (a[g(t)]) l/e> (x'[g(t)])
x(t)
x[g(t)]
for t > g( t)
a(td
x[g(t)]
[t - get)]
(3.13.7)
T. Similarly,
a- l /e>(t2) (a l /e>(t 2)x'(t2)) [get) - T]
x[g(t)]- x(T)
~ (a~~;:?])
1/0
x' [g(t)][g(t) _ TJ,
for some t2 E (T,g(t)). Therefore,
x[g(t)]
~ (a~~t~?])
1/0
x' [g(t)][g(t) - T].
It follows from x'(t) > 0 that
x[g(t)]
X'[g(t)]
(a[g(t)]) 1/0 [get) _ T] = (a[g(t)]) 1/0 k(t)g(t),

a(t2)
a(t2)
where k(t) = [get) - TJ/g(t). Since limHoog(t) = 00, it is clear that

limt->oo k(t) = 1. Thus, for each kl E (0,1), there exists Tkl ~ T such
that k(t) ~ kl for t ~ T kl . Then, we have
x[g(t)]
(a[g(t)]) 1/0
X'[g(t)] ~
a(t2)
klg(t)
for
By (3.13.7) and (3.13.8), we find that for t > get)
x(t)
x[g(t)]
(3.13.8)
t ~ T kl
~
Tkl
T,
< 1 + (a[g(t)]) l/e> (x'[g(t)]) [t _ get)]

a(tl)
<
x[g(t)]
1 + (a(t 2))1/0 (t - get))
a(tl)
klg(t)
< (a(t2))1/e> (_t_)

-
a(tl)
klg(t)
'
or
(3.13.9)
Oscillation and nonoscillation of haH-linear diHerential equations
221
Now, (3.13.4) and (3.13.5) easily follow from (3.13.9).

(II) Since x'(t) > 0 and a(t)(x'(t))<> is decreasing for t ~ T,
that a1 /<>(s)x'(s) ~ a1 /<> (t)x'(t) for t ~ s ~ T, and hence
x'(s) ~ a- 1 /<>(s)a 1 /<>(t)x'(t).
we find
(3.13.10)
Integrating (3.13.10) from T to t, we obtain
x(t)
x(t) - x(T)
(a 1 /O!(t)X'(t)).i a- 1 /<>(s)ds,
or x(t) ~ a1/Q(t)R[t, T]x'(t) for t ~ T. This completes the proof.
Lemma 3.13.2. Let x(t) E C 1 ([to, 00), JR+) be such that a1 / Q(t)x'(t) E
C1([to,00),JR), where a and a(t) satisfy conditions (i) and (ii) respectively, and (a(t)lx'(t)IQ-l x'(t))' ~ 0 for t ~ T ~ to. Then,
(II) when x'(t) < 0,

(3.13.11)
(b) when x'(t) > 0,
X(T) ~ al /<>(O')R[T,O']x'(O')
T ~ a ~ T.
(3.13.12)
Proof. (11) Since
x(O')
x(O') - X(T) > a1/<>(T)
(iT a-1/<>(S)dS) X'(T)
inequality (3.13.11) follows.
(12) Since
X(T)
X(T) - x(O') > a1/Q(O')
inequality (3.13.12) holds.
(iT a-1/Q(S)dS) x'(O')
Lemma 3.13.3. Consider the differential inequality
(a(t)lz'(t)IQ-1z'(t))'
+ q(t)lz[g(t)]I!3- 1z[g(t)]
::; 0,
(3.13.13)
where conditions (i) - (iv) hold and (3 > 0 is a constant. Let z(t)
be an eventually positive solution of (3.13.13). Then for some to and
0< Xo ::; z(to) there exists a solution x(t) of
(a(t)lx'(t)IQ-1x'(t))' + q(t)lx[g(t)]I!3- 1X[g(t)] = 0
(3.13.14)
Chapter 3
222
with x(to)
= Xo
such that 0 < X(i)(t) ~ z(i)(t), i
= 0, 1
for all t::::: to.
Proof. Since z(t) > 0 and (a(t)Jz'(t)J",-l z'(t))' ~ -q(t)zi3[g(t)] < 0

for all large t it follows that z'(t) is eventually of one sign, i.e., either
x'(t) > 0 for t::::: to, or there exists a tl ::::: to such that z'(t) < 0
for t::::: tl. Suppose Z'(t) < 0 for t::::: iI. Then since a(t)(z'(t))<> =
-a(t)(-x'(t))"', we find that - a(t)(-z'(t))<> ~ -a(td(-z'(td)<> for
t::::: tl, or equivalently, al/<>(t)z'(t) ~ al/<>(tl)z'(td, and hence
Z'(t) ~ (a1/<>(tl)z'(td) a-l/"(t)
t::::: tl.
for
(3.13.15)
Integrating (3.13.15) from iI to t, we get
which contradicts the fact that z(t) > 0 eventually. Thus, z'(t) > 0
eventually. Now, let Xo be such that 0 < Xo < z(T) for some T::::: to.
Then integrating (3.13.13) from t to u with u::::: t ::::: T and letting
u -+ 00, we obtain for t::::: T,
Z'(t) ::::: a-l/"'(t)
it{ex> q(s)zi3[g(s)]ds ) 1/<>
(3.13.16)
t::::: T.
(3.13.17)
t::::: T, we obtain
Integrating (3.13.16) from T to
z(t) ::::: z(T) +
(t,z).
(s,z)ds
z(T) + \[J(t,z),
Now, define
z(t)
Xo + \[J(t, x n ),
1,2" ...
(3.13.18)
Then from (3.13.17), we get by induction that 0 < xn(t) ~ z(t) for
= 0,1,2""
and Xn+l(t) ~ xn(t) for t::::: T, n = 0,1,2,.
Consequently, letting limn-+ex> xn(t) = x(t) and applying Lebesgue's
monotone convergence theorem, we find x(t) = Xo + \[J(t, x). Now, it
follows easily that x(t) has the desired properties. It is also clear that a
corresponding result holds for the negative solution z(t) of (2.13.13).
t::::: T, n
Lemma 3.13.4. Let q(t) E C([to, (0), JR+), T(t) E Cl([to, (0), JR), T(t) ~
t and T'(t)::::: 0 for t::::: to and limHex> T(t) = 00. If
liminf
t-+ex>
q(s)ds > -,
T(t)
e
223
(al) then the inequality x'(t)

tually positive solution,
+ q(t)X[T(t)]
::; 0 eventually, has no even-
(a2) then the inequality x'(t)

tually negative solution,
+ q(t)X[T(t)]
(a3)
then the equation x'(t)
+ q(t)X[T(t)] = 0
0 eventually, has no evenis oscillatory.
Lemma 3.13.5. Let q(t) E C([to, 00), lR+), T(t) E C1([to, 00), lR), T(t) ~
~ 0 for t ~ to. If
t and T'(t)
liminf
t-+oo
(t)
q(s)ds > -,
e
(b l ) then the inequality x'(t) - q(t)X[T(t)] ~ 0 eventually, has no eventually positive solution,
(b 2) then the inequality x'(t) - q(t)X[T(t)] ::; 0 eventually, has no eventually negative solution,
(b3)
then the equation x'(t) - q(t)X[T(t)]
= 0 is oscillatory.
Lemma 3.13.6. If y(t) is a positive and strictly decreasing solution of

the integral inequality y(t) ~ OO H(s, y[g(s)])ds, where HE C([to, 00) x
lR, lR), xH(t, x) > 0 for x -# 0, t ~ to ~ 0 and increasing in the second
variable, 9 E C([to, 00), lR), g(t) < t and limHoo g(t) = 00, then there
exists a positive solution x(t) of the equation x'(t)+H(t, x [g(t)]) = 0 such
that x(t) ::; y(t) eventually, and satisfies limHoo x(t) = 0 monotonically.
It

We are now in the position to state and prove the following result.
Theorem 3.13.1. Let conditions (i) - (iv) hold, and g(t)::; t. If for all
large T ~ To ~ to so that g(t) ~ To, t ~ T the first order delay equation
y'(t)
+ q(t)R!3[g(t), To] Iy[g(t)] lf3!a
sgn y[g(t)]
= 0,
t~T
(3.13.19)
is oscillatory, then equation (3.13.1) is oscillatory.

Proof.

~ to ~ o. As in Lemma 3.13.3 there exists a tl ~ to such
x(t) > 0 for t
that
x'(t) > 0
for
t ~ tl
(3.13.20)
and by Lemma 3.13.1(II), there exists a t2 ~ tl, g(t) ~ tl for t ~ t2 and
x[g(t)] ~ al/a[g(t)]R[g(t), h]x'[g(t)]
for
t ~ t2.
(3.13.21)
Chapter 3
224
Using (3.13.21) in equation (3.13.1), we obtain
+ q( t)R{3[g( t), tI] (alit> [g( t) ]x'[g( t)]) (3

:s; (a(t)(x'(t))OO)' + q(t):lP[g(t)] = 0 for t:::: t2'
(a( t)(x' (t) )00)'
Setting u:(t)
= a(t)(x'(t))c>,
t:::: t2
in the above inequa1ity, we get

(3.13.22)
Integrating (3.13.22) from
t:::: t2
to
v,
and letting
v,
-+
00,
we find
Clearly, the function w(t) is strictly decreasing for t:::: t2' Hence, by
Lemma 3.13.6, there exists a positive solution y(t) of equation (3.13.19)
with limHoo y(t) = O. But this contradicts the assumption that (3.13.19)
is oscillatory. This completes the proof.
Lemma 3.13.4 when applied to (3.13.19) leads to the following corollary.

Corollary 3.13.1. If for every T:::: To:::: to, t :::: g(t) :::: To,
t q(s)RC>[g(s), To]ds
} get)
liminf
t-+oo
or
JOO q( S )ROO [g( s), To]ds
> -
00
when
when
t:::: T,
0'= (3,
0'< {3,

Theorem 3.13.2. Let conditions (i) - (iv) hold, 0'= (3, g(t) > 0 and
g'(t) > 0 for t:::: to. If there exists a function p(t) E CI([to,oo),JR+)
such that
li~~p
where,),
t [
to p(s)q(s)
1/(0'
+ l)(a+I),
(p'(S))c>+I ]
-')'a[g(s)] (p(s)g'(s))a ds
00,
(3.13.23)

x(t) > 0 for t:::: to :::: o. As in Theorem 3.13.1 there exists a tl :::: to such
Proof.
that for
t:::: t I ,
x'(t) > 0
and
a(t)(x'(t))a:s; a[g(t)](x'[g(t)])a.
(3.13.24)
225
Define w(t) = p(t)a(t)(x'(t))"'/x13 [g(t)] for t ~ tl. Then for t ~ tl,
w'(t) = - p(t)q(t) + p'(t) w(t) _ (3p(t)g'(t) a(t)(x'(t))"'x'[g(t)]. (3.13.25)

p(t)
x 13 + 1 [g(t)]
Using (3.13.24) in (3.13.25) and putting a = (3, we get for t
p'(t)
,
( a(t)
w'(t) < -p(t)q(t)+ p(t) w(t)-ap(t)g (t)a(t) a[g(t)]
t1,
)1/'" (x[g(t)]
x'(t) )"'+1
,
or
w'(t) < - p(t)q(t)
+ p'(t) w(t) p(t)
ag'(t)(a[g(t)]p(t))-l/"'W(",+1)/", (t).
(3.13.26)
Set
'( ))"'/("'+1)
ag t
w(t)
(a[g(t)]p(t))1/("'+l) '
>.=a+1>1
a
and
B = (_a_) '" [p'(t) (a[g(t)]p(t))l/(a+1)(a g'(t))-"'/("'+l)] '"

a +1
p(t)
in Lemma 3.2.1, to obtain
p'(t) w(t) _ ag'(t)(a[g(t)]p(t))-l/"'w(a+1)/"'(t)

p(t)
::;
+1
)"'+1 [a[g(t)] (p(t)g'(t))'"

(p'(t))"'+1 ]
for
t~tl'
Thus, inequality (3.13.26) gives
,
W (t) ::; - p(t)q(t)
(p'(t))"'+l
+ ,a[g(t)] (p(t)g'(t))'"
for
t ~ tl
Integrating the above inequality from tl to t, we get
Taking lim sup on both sides of (3.13.27) as t --+ 00, we find a contradiction to condition (3.13.23). This completes the proof.
When a
result.
= {3 = 1, Theorem 3.13.2 reduces to the following well-known
226
Chapter 3
Corollary 3.13.2. Let conditions (i) - (iv) hold, a = (3 = I, g(t) > 0

and g'(t) >0 for t::;>to. If there exists a function p(t) EC 1 ([to,oo),lR+)
such that
lim sup
t ...... oo
t [
to
1
(p'(S))2 ]
p(s)q(s) - -a[g(s)] ( ) '() ds
4
psg s
00,

Although, the following corollary is immediate, we shall prove it without
using Lemma 3.2.1.
Corollary 3.13.3. Let condition (3.13.23) in Theorem 3.13.2 be replaced
by
limsup
1. ...... 00
and
lim
t ...... oo
to
r' p(s)q(s)ds
(3.13.28)
00
lto
(p'(S))Q+l
a[g(s)] (psg
( ) '( S )) ds <
(3.13.29)
00,
then the conclusion of Theorem 3.13.2 holds.

x(t) > 0 for t::;> to ::;> O. We define the function w(t) as in Theorem
3.13.2 and obtain (3.13.26) for t::;> tl.
Integrating (3.13.26) from tl to t,
w(t)
It
+ It
p(s)q(s)ds
t,
-It
t,
we find
p'((s)) w(s)do5
P
05
(3.13.30)
g'(s)(a[g(s)]p(s))-l/QwCQ+ll/Qds < w(tr) = c,
t,
where c > 0 is a constant.
ft';' p' (s )w( s) / p( s )ds < 00. Then it follows from

(3.13.30) that ftt, p(s)q(s)ds::; c+ ftt, p'(s)w(s)/p(s)ds, which as t --t 00
gives ft';' p(s)q(s)ds < 00. But, this in view of (3.13.28) is impossible.
Suppose first that
Suppose next that
00
t,
It
p'(s)
-(-) w(05)do5 =
Ps
Then by (3.13.30), we obtain
t,
p(s)q(s)ds::; c+
t,
p'(s)
-(-) w(s)ds-a
PS
It
t,
(3.13.31)
00.
u+1
g'(s)(a[g(s)]p(s))-;;;w~(s)ds.
(3.13.32)
227
Now, estimating the second integral in (3.13.32) by Holder's inequality, we

have
t,
p'(S) w(s)ds =
p(s)
It
(pl(S) (a[g(s)]p(s))1/(a+1)(agl (s))-a/(O+1))

pes)
t,
((a[g(s)]p(s))-1/(a+l)(ag' (s))a/(a+1)w(s)) ds
(I,
I, (:((;;)
a/(o+l)
<
agl(s)(a[g(S)jp(s))-1/O'u/a+1)/ads)
I
<>+1
) 1/(0+1)
a[g(s)jp(s)(ag'(s))-O'ds
(3.13.33)
(p'(s)w(s)jp(s))ds --+ 00 as t --+ 00, we
Since (3.13.31) implies that
find from (3.13.29) and (3.13.33) that there exists a t2 2': tl such that for
itt,
t 2': t2,
It
t,
p'(S) w(s)ds :s; a

pes)
It
g'(s)(a[g(s)jp(s))-1/0w(o+1)/a(s)ds.
t,
Using this in (3.13.32), we conclude that limsuPt->oo

J/, p(s)q(s)ds :s; c,

1
(a(t)lx'(t)la- X'(t))
k
+ ta+1Ix[Atjla-1x[Atj
=
0,
t>0
(3.13.34)
where a, A, k are positive constants, 0 < A :s; 1 and aCt) E C([to, 00),
lR +). It is easy to verify the following:
(AI) If aCt) = 1, pet)
t"'+l and
k >
1 (
A'"
+1
",+1
'
then all conditions of Theorem 3.13.2 are satisfied, and hence equation
(A 2 ) If aCt) = 1jtf3 where {J > 0 is a constant, then for all (J, A and
k the hypotheses of Corollary 3.13.2 are satisfied with pet) = t a +!, and
therefore, equation (3.13.34) is oscillatory.
Now, for each t 2': to, we define ,et)
,et) 2': t and go ,et) = t.
sup{s 2': to: g(s):S; t}. Clearly,
Theorem 3.13.3. Let conditions (i) - (iv) hold, and a = (J, get) :s; t for
t 2': to. Then equation (3.13.1) is oscillatory if either one of the following
conditions holds
228
Chapter 3
(C 1 ) when a'(t):::::: 0 then for t 2 T 2 to,
(3.3.35)
(C 2 ) when a'(t) 2 0 then for t 2 T 2 to,

limsup R<>[t, T1lOO (g(S)<> q(s)ds > 1,
t---+oo
t
s
(3.13.36)
(C3 ) for t 2 T 2 to,

lim sup R<> [t, T1lOO q( s )ds > 1.
,(t)
t---+oo
(3.13.37)

x(t) > 0 for t 2 to. As in Lemma 3.13.3, there exists a t1 2 to such that
x'(t) > 0 for t 2 it and by Lemma 3.13.1(II) there exists a t2 2 t1 such
Proof.
that
(3.13.38)
Integrating (3.13.1) from t 2 t2 to u and letting u --+ 00, we find
a(t)(x'(t<> 2
00
q(s)xc<[g(s)]ds.
(3.13.39)
Using (3.13.38) in (3.13.39), we get for t 2 t2,
By Lemma 3.13.1(1), for each constant k E (0,1) there exists Tk 2 t1

such that
xQ[g(t)] 2 kQ (:s:)
or
x<>[g(t) 2 k<>
(g~t) QxQ(t)
(g~t)
x<>(t)
when a'(t) :::::: 0 and t 2 Tko
when a'(t) 20 and t 2 Tk .
Thus, from (3.13.40) for t 2 T = max{Tk' t2},

t 2 T, we obtain
when a'(t):::::: 0 and
(3.13.41)
229
or when a'(t) ~ 0 and t ~ T,

1
~ kO< Ra[t, t2JlO (g~s))
0<
q(s)ds.

lim sup RO< [t, t2J roo a(( s )) (g( S))" q( s )ds
t-HYO
a t1
s
it
or
lim sup R"'[t, t2J roo (g(s))O< q(s)ds

t-too
s
it
= b2 <
00.
Suppose condition (3.13.36) holds, then there exists a sequence

such that limn-too Sn = 00 and for each fixed t2:::; S1,
{sn}~=1
(3.13.42)
For
b2 ; 1
(b 2
1)/2> 0, there exists an integer N > 0 such that
= b2-E < Ra[sn, t2Jl~ (g~s)) '" q(s)ds
Choose k1 such that (2/(b 2

(3.13.43), we have
1
+ 1))1/'" <
~ kfR"'[sn,t2Jl~(g~S))a q(s)ds
k1
>
<
for n
~ N.
(3.13.43)
1. Then by (3.13.42) and
(b2~1)
C2;1)
for all sufficiently large n. This contradiction shows that condition (3.13.36)
does not hold. The other case can be considered similarly and hence omitted.
Next, by 'Y(t}
xa(t)
t and (3.13.40), we have

Ra[t, t2Jjt q(s)x"[g(s}Jds for
'Y(t)
~ t2'
Since x(t) is increasing and g(s) ~ t for s ~ 'Y(t), it follows that

1 ~ Ra[t, t2J J'YCt) q(s)ds, which contradicts condition (3.13.37).
Corollary 3.13.4.
t ~ to and
Let conditions (i) - (iv) hold, a: = (3, g(t) :::; t for
a(t) == 1. If either
limsup t a roo (g(S))a q(s)ds > 1,

t-too
s
it
230
Chapter 3
or
limsup to.
t--+CXl
CXl
q(s)ds > 1,
let)
holds, then equation (3.13.1) is oscillatory.

Example 3.13.3. Consider the
ha.lf~linear
(t<>lx'(t)la~lX'(t))' + t(lne~)a+1 Ix [~] 10.-1 [~] = 0, t ~ T = 1 ~ to> 0,

X
(3.13.44)
where a and c are positive constants. Let R[t,T] = J;a~l/<>(s)ds =
J: ds/s = Int, so that

R a [t, T]
1 (g(
CXl
s) '" q( s )ds = (In t) a

-s_ (lnt)a
1 (1) '"
00
(~)a
2
"2
c )0.+1 ds
s(ln es
2~ac (~)a
c
_
a(lnet)<> -
Inet
Taking lim sup in the above equality as t -+ 00, we find in view of

Theorem 3.13.3 that equation (3.13.44) is oscillatory provided c> a(2"').
Theorem 3.13.4. Let conditions (i) ~ (iv) hold, a > 1, f3 > 1,

<
g(t) :s; t and g'(t) > 0 for t ~ to, and assume that there exists a
function p(t) E C 1([to,oo),lR+) such that
p'(t) ~ 0,
k(t)
a1 /<>[g(t)]p'(t)
g'(t)
and k'(t):s;
for t ~ to. (3.13.45)
If
JCXl
p(s)q(s)ds
(3.13.46)
00,

x(t) >
for t ~ to ~ 0. As in Theorem 3.13.2, there exists a t1 ~ to
such that (3.13.24) holds. Define the function w(t) as in Theorem 3.13.2
to obtain (3.13.25) which takes the form
w'(t) :s; - p(t)q(t) + p'(t) a(t)(x'(ta

xP[g(t)]
for
t;:::: t2
~ t1.
(3.13.47)
There exist a constant b> 0 and a t3 ;:::: t2 such that (a1/<>(t)x'(t)"'-1:s;

b for t ~ t3. Thus, (3.13.47) becomes
a1/<> (t)x'(t)
w'(t) < - p(t)q(t) + b xP[g(t)]
for
t;:::: t3.
(3.13.48)
231
Next, there exists a T ~ t3 such that al/ct)x'(t)::; al/c<[g(t)]x'[g(t)] for
t ~ T and hence
w'(t) ::; - p(t)q(t) +b (a1/c<[g(t)]p'(t)) (x'[g(t)]g'(t))

x!3[g(t)]
g'(t)
But, by the Bonnet theorem for a fixed t
have
t
iT
for
t ~ T.
T and for some ~ E [T, t], we
(al/c<[g(s)]p'(s)) (x'[g(s)]g'(S)) d
g'(s)
x!3[g(s)]
,s
( a 1/C<9[(T)]P'(T)) (f. x' [o9( s )]09' (s) ds
g'(T)
iT x!3[g(s)]
[g(f.)]
k(T)
k(T)
u-!3du = - - [x-!3+ l [g(T)]_
(3 - 1
x[g(T)]
::; (3k(T) x 1-!3[g(T)]

-1
== K <
x-!3+1[g(~)ll
(Xl,
where K is a positive constant. Thus, we find
(t (a1/c<[g(S)]P'(S)) (x'[g(s)]g'(S)) ds < K
iT
g'(s)
x!3[g(s)]
t ~ T.
for
(3.13.49)
J;
Therefore, in view of (3.13.49), it follows that

p(s)q(s)ds ::; -w(t)+w(T)
+bK < (Xl, which contradicts (3.13.46). This completes the proof.
Theorem 3.13.5. Let conditions (i) - (iv) hold, (3 > a, g(t) ::; t and
~ 0 for t ~ to. If
g'(t)
00
a-l/c<[g(s)]g'(s)
(1
00
q(U)dU) l/c< ds
(3.13.50)
(Xl,

x(t) > 0 for t ~ to ~ O. As in Theorem 3.13.4, we take p(t) = 1 to
obtain ft q(s)ds ::; a(t)(x'(t))c< /x!3[g(t)] for t ~ tl ~ to, or
Proof.
OO
a-1/C<[g(t)]g'(t)
(1
00
q(S)dS) 1/c< ::;
x~~/~1~~~~~)
for
~T ~t
l .
Integrating the above inequality from T to t, we find
[g(t)]
v-!3/udv
x[g(T)]
< _a_ x l-(!3/c<) [g(T)] < (Xl,

{3-a
232
Chapter 3
which contradicts condition (3.13.50) and completes the proof.

0,
where (3 > et > 0 are constants. Since,
/t
a -1/'" [g(s)]g'(s)
(100)
s
q(u)du
Jt
s (
I/n
1 ) ds
-~
4ets 2
ds ~
/t ~ (/00
Js
= -1 In t
Set
t >0
V.- I - 2Ct dv.
-.-t 00 as
(3.13.51)
1/",
ds
t -+ 00
it follows from Theorem 3.13.5 that equation (3.13.51) is oscillatory.
3.13.3. Some Extensions and More Oscillation Criteria

In this subsection we begin with the more general equation
(a(t)lx'(t)I"-l X '(t)'
+ q(t)F(x[g(t)]) =
(3.13.52)
0,
where et, a, g, q satisfy conditions (i) - (iv), and
FE C(IR, IR) and xF(x) > 0 for x i= O.
(v)
Our interest here is to obtain oscillation criteria for (3.13.52) similar to

those presented in Subsection 3.13.2 without assuming that the function
F is monotonic. For this, we need the following notation and lemma. Let
(-00, -to]
IRto
(-00,0) U (0,00)
C(IR)
{j
to > 0
if to = 0,
[to, (0)
if
C(IR, IR) and xf(x) > 0 for x
i= O}
and
{f E C(IR): f is of bounded variation on every interval

[a, b] ~ IRto} .
Lemma 3.13.7. Suppose to> 0 and f E C(IR). Then, f E CB(IR to )
if and only if f(x) = G(x)H(x) for all x E IR to ' where G: IRto -+ IR+
is nondecreasing on (-00, -to) and nonincreasing on (to, (0), and H:
IRto -+ IR is nondecreasing on IRto'
Now, we shall prove the following comparison theorem.
233
Theorem 3.13.6. Assume that FE C(lRto)' to > 0 and let G and H

be a pair of continuous components of F with H being the nondecreasing
one. Moreover, assume that there exists a positive constant (3 such that
H(x) sgn x 2 Ixl i3
for
(3.13.53)
x:f. 0,
conditions (i) - (v) hold, get) :S; t for t 2 to. If for every constant k > 0
and all large T 2 To 2 to, get) 2 To for t 2 T, the equation
(a(t)lx'(t)IQ-1x'(t)'
+ q(t)G(kR[g(t), ToDlx[g(t)]I i3 - 1x[g(t)] = 0
(3.13.54)
is oscillatory, then (3.13.52) is also oscillatory.
Proof.
Let x(t) be a nonoscillatory solution of equation (3.13.52).
Without loss of generality assume that x(t) > 0 for t 2 to > o. As in
the proofs given earlier it is easy to see that x'(t) > 0 and a(t)(x'(tQ
is nonincreasing for t 2 h 2 to. Thus, there exists a t2 2 tl and a
constant b1 > 0 such that x'(t) :S; b1a-1jct) for t 2 t 2. Integrating this
inequality from t2 to t, we conclude that there exist a constant b > 0
and a t3 2 t2 such that
x[g(t)] :S; bR[g(t) , t2]
for
t 2 t 2
(3.13.55)
Now, it follows from equation (3.13.52), (3.13.53) and (3.13.55) that
+ q(t)G(bR[g(t), t2])X i3 [g(t)]

(a(t)(x'(tQ)' + q(t)G(x[g(t)])H(x[g(t)]) = o.
(a(t)(x'(tc<)'
:S;
(3.13.56)
But, in view of Lemma 3.13.3, equation (3.13.54) has an eventually positive
solution, which is a contradiction. This completes the proof.

We note that Theorem 3.13.6 together with the results presented earlier
can be applied to equations of the type (3.13.52) with F being anyone of
the following functions.
F(x) = Ixl i3 - 1xexp(-lxl')"), (3 and 'Yare positive constants,

F(x) = Ixl i3 - 1x sech x, (3 > 0 is a constant,
Ixl,6-1x
F(x) -- 1 + Ixl')" , (3 and '"I are positive constants.

Next, we shall extend the results of Section 3.13.2 to equations of the
neutral type of the form
(a(t) I(x(t)
+ p(t)x[r(t)])'I Q- 1 (x(t) + p(t)x[r(t)]))' +q(t)lx[g(t)]I i3 - 1x[g(t)]

=
0,
(3.13.57)
Chapter 3
234
where a, (3, a, 9 and q satisfy conditions (i) - (iv), and
p(t) E C([to,oo),lRo),
(vii) T(t) E C([to, 00), lR), T'(t) > 0 and limt--+oo T(t) = 00.
(vi)
For equation (3.13.57) we shall prove the following comparison results.

Theorem 3.13.7. Let 0::::: p(t) ::::: 1, T(t) < t and g(t) ::::: t for t::::: to,
and p(t) of:. 1 for all large t. If the equation
(a(t)lz'(t)la-IZ'(t)'
+ q(t)(l - p[g(t)])!3lz[g(t)]I!3- I Z[g(t)]
(3.13.58)

Theorem 3.13.8. Let p(t) ::::: 1, T(t) > t, a(t) = T- I 0 g(t) ::::: t and
a'(t) ::::: 0 for t::::: to, and p(t) of:. 1 for all large t. If the equation
(a(t)lw'(t)la-lW'(t)'
+ q(t)(P[g(t)])!3lw[a(t)]I!3- l w [a(t)]
0 (3.13.59)
and
T- I
is the inverse function of
T,
Proof of Theorems 3.13.7 and 3.13.8. Let x(t) be a nonoscillatory

solution of equation (3.13.57), say, x(t) > 0 for t::::: to ::::: O. Define
y(t) = x(t) + p(t)X[T(t)]. Then, we have
(a(t)ly'(t)I"-ly'(t)'
+ q(t)lx[g(t)]I!3- l x [g(t)]
O.
(3.13.60)
It is easy to verify that y(t) > 0 and y'(t) > 0 for t::::: tl ::::: to. Now
using the hypotheses of Theorem 3.13.7, we find
x(t)
>
y(t) - p(t)X[T(t)]
y(t) - p(t)[Y[T(t)]- p[T(t)]X[T 0 T(t)]]
y(t) - p(t)Y[T(t)] > (1- p(ty(t) for t::::: T::::: tl.
(3.13.61)
(a(t)(Y'(t")'
+ q(t)(l - p[g(t)]ly!3[g(t)] :::::
for
t::::: T.
By Lemma 3.13.3, equation (3.13.58) has an eventually positive solution,

235
Next, we use the hypotheses of Theorem 3.13.8, to obtain

1
x(t)
p[rl(t)] (Y[T-l(t)]_ X[T-l(t)])

1
(Y[T- 1 0 T-1(t)]
X[T- 1 0 T-1(t)])
Y[T-1(t)]
p[rl(t)] p[rl(t)] p[T- 1 0 r1(t)] p[T- 1 0 r1(t)]
> Y[T-1(t)]
Y[T- 1 0 T-1(t)]
p[rl(t)] p[r1(t)]p[r 1 0 T-1(t)]
2: P(t)Y[T-1(t)] for t 2: r
for some T* 2: it.
(3.13.62)
Using (3.13.62) in equation (3.13.60), we find
(a(t)(y'(t"')'
+ q(t) (P[g(t)])i3 yi3 [O"(t)]
S 0 for
2: T*.
Once again, by Lemma 3.13.3, we arrive at the desired contradiction. This

completes the proof.
Example 3.13.5. The neutral differential equations
(3.13.63)
and
( to.l(x(t)
+ 2x[c*t])'1 '" - 1 (x(t) + 2x[c*t])' ) ' + q2(t)lx[cit]I i3 - 1 x[cit]
0,
(3.13.64)
where o!, /3, c, c*, Cl and ci are positive constants such that Cl <
1, ci > 1, ci S c* and ql (t), q2(t) E C([to, 00), lR+) are oscillatory if the
associated differential equations
=
(3.13.65)
and
=
(3.13.66)
are oscillatory. Further, the oscillation of (3.13.65) and (3.13.66) can be

discussed by applying the results of Section 3.13.2. The statements of such
results are left to the reader.
Finally, we shall consider the damped differential equation
(a(t)lx'(t)lo.-1x'(t))'
+ f(t, x[t -
T], x'[t - 0"])
0,
(3.13.67)
Chapter 3
236
where Q and a satisfy conditions (i) and (ii) respectively,
rea.l constants, and FE C([to, oo) x m?,lR).
We shall assume that there exist positive constants
function q(t) E C([to,oo),lR+) such that
J(t, x, y) sgn x ~ q(t) Ixl,6lyll-t
for
xy
-=1=
and rr are
f3 and
0, t ~ to.
{L
and a
(3.13.68)
Theorem 3.13.9. Let condition (3.13.68) hold, {L:5. Q, T be any real

number, and rr ~ O. If for every positive constant B, the delay equation
y'(t) + Ba-I-'/<>[t - rr]q(t)ly[t - rr]II-'/C< sgn y[t - rr] = 0
(3.13.69)
is oscillatory, then (3.13.67) is also oscillatory.

x(t) > 0 for t ~ to ~ O. It is easy to verify that x'(t) > 0 for t ~ tl ~ to.
There exist a positive constant c and a t2 ~ tl such that
x,6[t - T] ~ c
for
t ~ t2.
(3.13.70)
Using condition (3.13.68) and (3.13.70) in (3.13.67), we get (a(t)(x'(t))<' +

cq(t)(x'[t - rr])1-' S; 0 for t ~ t2. Setting z(t) = (a1/<>(t)x'(t))<> for
t ~ t2, we find z'(t) + cq(t)a-I-'/<>[t - rr]zl-'/<>[t - rr] :5. 0 for t ~ t 2.
Integrating this inequality from t ~ t2 to u and letting u ~ 00,
we obtain z(t) ~ CftOO q(s)a-I-'/<>[s - rr]zl-'/<>[s - rr]ds, t ~ t2. But, in
view of Lemma 3.13.6, equation (3.13.69) has a positive decreasing solution
yet), limt--+oo yet) = 0, which is a contradiction.
Theorem 3.13.10.
Let condition (3.13.68) hold, f3 + {L :5.
rr ~ o. If for all large T, t ~ T + T, the delay equation
Q,
and
T ~
v'(t)+a(t)R,6[t-T, T]a-I-'/<> [t-rr] Iv[t-rr] 1(,6+1-')/<> sgn v[t-rr] = 0 (3.13.71)

x(t) > 0 for t ~ to. It is easy to verify that x'(t) > 0 for t ~ tl ~ to
and by Lemma 3.13.1(II), there exists a T ~ tl such that
x[t - T]
> R[t - T, T]a1/<>[t - T]X'[t - T]

> R[t - T, T]a1/<>[t - rr]x'[t - rr] for t
~ T
+ T.
(3.13.72)
Using condition (3.13.68) and (3.13.72) in equation (3.13.67), we get
(a(t)(x'(t))<' +q(t)R,6[t-T, T]a,6/<> [t-rr](x'[t-rr]),6+1-' < 0 for t ~ T+T.

Setting yet) = a(t)(x'(t))<>, t
y'(t)
+ q(t)~[t -
+ T,
we find
T, T]a-I-'/<>[t - rr]y(,6+I-')/<>[t - rr] < 0
for
t ~T
+ T.
237
The rest of the proof is similar to that of Theorem 3.13.9 and hence omitted.
Remark 3.13.1. The results similar to that of Corollary 3.13.1 can be

easily stated from Theorem 3.13.9 and 3.13.10.
Example 3.13.6. The damped differential equation
(t2lx'(t)lx'(t))'
+ q(t)lx'[t -
Tllx[t - T] = 0,
t >
(3.13.73)
is a constant, and q(t) E C([to, 00), m+) is oscillatory by

where T >
Theorem 3.13.10 provided for T = 1 ::;, to, the equation
y'(t)
+ In(t t-
T) q(t)y[t - T] = 0,
t> 1 + T
(3.13.74)
is oscillatory. We also note that (3.13.74) is oscillatory if

liminf
t--+oo
t-T
(In(S-T)) q(s)ds
.
S -
1
e
> -
(3.13.75)
(ef. Lemma 3.13.4(303))' Therefore, (3.13.73) is oscillatory if condition

(3.13.75) holds.
3.13.4. Classification of Nonoscillatory Solutions

Here, we shall consider the following second order functional differential
equation
(a(t)1j;(X'(t)))' + F(t, x [g(t)]) = 0,
(3.13.76)
where
art) E C([to, 00), lR+),

(ii)
1j; E C(lR, lR), 'I/; is strictly increasing, sgn'l/;(x) = sgn x and
'I/;(lR) = lR,
(iii) g(t) E C([to,oo),lR), g'(t)::;, 0, g(t)::::: t and limHoog(t) = 00,
(iv)
F E C([to,oo) x lR,lR), F(t,x) is nondecreasing in the second
variable, and sgn F(t, x) = sgn x for each fixed t::;, to ::;, 0.
(i)
In what follows, we shall assume that
10= 1'1/;-1 (a~sJ 1ds

where '1/;-1
notation
: lR
00 for every constant k j 0, to::;'
--+ lR denotes the inverse function of 1j;,
(3.13.77)
and use the
(3.13.78)
Chapter 3
238
From (3.13.77) and (3.13.78) it follows that Wk ,T(a; T) = 0, tlim

IWk T(a; t)1
-H)Q'
= 00 for every k -=I 0, IWk,T(a; t)1 > IWe,T(a; t)l, t > T for Ikl > 11
with k > 0 and limk ..... o Wk,T(a; t) = 0 for all t 2: T.
We begin by classifying nonoscillatory solutions of (3.13.76) according

to their asymptotic behavior as t --+ 00.
Lemma 3.13.8. Any nonoscillatory solution x(t) of equation (3.13.76)
is one of the following three types
limHoo a(t)1jJ(X'(t = constant -=I 0,

limHoo a(t)'Ij;(x'(t = 0, limt ..... oo Ix(t)1 = 00,
(III) limHoo a(t)'Ij;(x'(t = 0, limHoo x(t) = constant -=I O.
(I)
(II)
Proof.
Let x(t) be a nonoscillatory solution of equation (3.13.76).
Without loss of generality, we can suppose that x(t) > 0 for t 2: to > O.
From equation (3.13.76), it follows that
(a(t)'Ij;(x'(t)' = - F(t, x [g(t)]) < 0 for t 2: tl 2: to

and so a(t)'Ij;(x'(t
is decreasing for t 2: tl'
We claim that a(t)'Ij;(x'(t > 0, t 2: tl so that limHoo a(t)'Ij;(x'(t

2: O. In fact, if a(t2)'Ij;(X'(t2 =-k < 0 for some t2 2: tl, then
a(t)'Ij;(x'(t ::::; -k, t 2: t2, which is equivalent to x'(t)::::; 1jJ-I (-k/a(t
for t 2: t2' Integrating this inequality from t2 to t and letting t --+ 00,
we find in view of (3.13.77) that x(t) --+ -00 as t --+ 00. But this
contradicts the positivity of x(t). Therefore, a(t)'Ij;(x'(t > 0 for t 2: tl'
A consequence of this observation is that x'(t) >
for t 2: t l . Thus,
the limit limHoo a( t )'Ij;( x' (t is either positive or zero. In the first case
x(t) is unbounded, since there are positive constants kl' k2 (k l < k 2)
and tl suchthat Wk,t,(a;t)::::;x(t)-X(tl)::::;Wk2,t,(a;t) for t2:tl' In
the second case, since x(t) is increasing, x(t) tends to a positive limit,
finite or infinite as t --+ 00. This completes the proof.
Now, we shall present criteria for the existence of nonoscillatory solutions of equation (3.13.76) of the types (I), (II) and (III).
Theorem 3.13.11. Suppose for each fixed k -=I 0 and T 2: to 2: 0,
w,T(a;g(t)
k>O Wk,T(a;g(t))
lim
40,
(3.13.79)
uniformly on any interval of the form [T* , 00) for some T* > T. Then
equation (3.13.76) has a nonoscillatory solution of the type (I), if and only
if,
roo W(s,cwk(a;g(s))lds
ito
<
00
(3.13.80)
Oscillation and non oscillation of baH-linear differential equations

for some constants k
=f. 0
and
239
> 0,
Proof (The 'only if' part). Let x(t) be a nonoscillatory solution of the
type (I) of equation (3.13.76). Without loss of generality, we assume that
x(t) > 0 eventually. There exist positive constants CI, kl and tl such
that x[g(t)] ~ cI'llkJ(a;g(t)) for t ~ t l . An integration of(3.13.76) yields
ItOO F(s,x[g(s)])ds < 00, which when combined with the earlier inequality
leads to
F(s, cl'llk J (a;g(s)))ds < 00.
It":
(The 'if' part). Suppose (3.13.80) holds for some constants c > 0 and
k > O. Because of (3.13.79) we can choose constants C> 0 and T ~ to > 0
so that C < k/2, T* = inft>T g(t) and
lOO F(s, 'lI2(a; g(s))ds

Define the subset
C([T*, (0)) by
{x
X =
C([T*, (0))
of
C([T*, (0)): 'lI,T(a; t)

for t 2 T,
x(t)
C.
(3.13.81)
W: X --+
and the mapping
x(t)
'lIu,T(a; t),
0 for T* ~ t ~ T}
~
(3.13.82)
and
Wx(t)
It'lj;-l [ats) (2C-l F(u,x[g(U)])dU)]dS,

S
0,
T*
T.
(3.13.83)
W maps X into itself. If x
(i)
o~
lS
F(u,x[g(u)])du
loo
X, then since
F(u, 'lI2(a;g(u))du < C,
we obtain from (3.13.83) that
'lj;-l (afs)) ds
Wx(t) <
'lj;-l
(a~!)) ds,
~T
~T
implying that W x E X.
(ii) W is continuous. Let {xn}~=l be a sequence in X converging to
x E X as n --+ 00 in the topology of C([T, (0)). Fix m E {l, 2, .. }
with m ~ T. The Lebesgue dominated convergence theorem implies that
F(S,Xn[g(s)])ds --+
F(s,x[g(s)])ds as n --+ 00, and so Wxn(t)--+
Wx(t) uniformly on [T,m]. Thus, WXn --+ Wx in C([T, (0)).
I;
I;
Chapter 3
240
(iii) W(X) is relatively compact. This follows from the relation
1rlla~) (2~ ltF(S,X[9(S)])ds)J
:<:::
(Wx)'(t) =
<
1rllatt) (2+ lt F (8,W 2c (a;g(S))dS))J,
t"2T
which holds for all x E X.

Now, the Schauder-Tychonov fixed point theorem guarantees that there
exists a x E X such that VV x = x. Differentiation of the integral equatioll
Wx(t) = x(t), t "2 T shows that x(t) is a positive solution of equation
(3.13.76) for t"2 T. It is clear that :r(t) is of the type (I).
In the case when (3.13.80) holds for some constants k < 0 and c > 0,
a similar argument can be used to construct a negative solution of the type
(I) of equation (3.13.76).
Theorem 3.13.12. Equation (3.13.76) has a nonoscillatory solution of

the type (III) if and only if
(3.13.84)
for some constant c =I- O.
Proof (The 'only if' part). Let :r(t) be a positive solution of the type
(III) of equation (3.13.76). Then there is a positive constant Cl and tl
such that x[g(t)] "2 Cl for t"2 tl' Integrating (3.13.76) from t"2 tl to
u and letting u --+ 00, we find a(t)1/;(x'(t)) = ItOO F(s, x [g(s)])ds for
t "2 t 1 , which implies that
x'(t) = 1/;-1 latt)
00
F(S,X[9(S)])dsJ
for
t"2
tl'
Integrating this equation again from tl to t and using x(t) "2

and letting t --+ 00, we obtain
t "2
tl
Cl
for
(The 'if' part). Suppose (3.13.84) holds. Let C> 0 be fixed arbitrarily
and take T"2 to > 0 so large that T* = inft2:T g(t), and
241
and define
{x
and
Vx(t) = {
C([T*, (0)) : c/2::::; x(t) ::::; c for
-1
t:2': T*}
~p-l [_(1) F(u, x[g(U)])dU] ds,

cas s
Vx(T) for T*::::; t ::::; T.
c
00
00
It is easy to verify that V is continuous and maps X

subset of X, and hence V has a fixed point x in X,
desired solution of the type (III) of equation (3.13.76).
into a compact
which gives the
Similarly, it can be shown that (3.13.76) possess a negative solution

of the type (III) if (3.13.84) holds for some c < O. This completes the
proof.
A characterization of the type (II) solutions of (3.13.76) is difficult.

However, we have the following result.
Theorem 3.13.13. Suppose (3.13.79) holds, and g(t) = t. Equation
(3.13.76) has a nonoscillatory solution x(t) with limHoo Ix(t)1 = 00 if
(3.13.80) holds for some k i- 0, c > 0 and
(3.13.85)
for every nonzero constant b such that kb
> O.
Proof. It suffices to consider the case where k > 0 and b > O. Let
d > 0 be an arbitrary fixed constant and choose t > 0 small enough and
T :2': to > 0 large enough so that d + W(ajt) ::::; cWk(ajt) for t:2': T,
and
F(s,d+w(ajs))ds::::; t. This is possible because of (3.13.79) and
the fact that limt ..... oo Wk(ajt) = 00. Now, consider the set Y and the
mapping U defined by
J;
{xEC([T*,oo)): d::::;x(t)::::;d+wt(ajt), t:2':T}
and
Ux(t)
d+
It ~-l
rats)
(t - lS
F(u, X(U))dU) ] ds
for
t:2': T.
By the Schauder-Tychonov fixed point theorem it follows that U has a

fixed element x E X. Now, x = x(t) is a solution of (3.13.76) follows
from the differentiation of the integral equation x(t) = Ux(t), t ~ T.
Chapter 3
242
From (3.13.85), we have
x(t)
2:
1>r
+
d+
> d
[ats)
'ljJ-I [ats)
> d+ it'ljJ-I [ats)
( -is F(u, X(U))dU) ] ds
l
l
co
co
F(u, X(U))dU] ds
F(u,d)dU] ds
We also note that limt-+co a(t)'ljJ(x'(t)) 2: O.
00
0,
as
00.
(lx'(t)I'" sgn x'(t))'
+ q(t)IX[At]li3 sgn X[At]
(3.13.86)
where a, (3, A are positive constants, 0 < A :s: 1 and q(t) E C(JR+, JR+).
Equation (3.13.86) is a special case of (3.13.76) in which a(t) = 1, 'ljJ(x) =
x"', g(t) = At, F(t,x) = q(t)x i3 , 'ljJ-l(x) = xli"', and 'l1 k ,T(a;t) =
kl/i3[t-T], t 2: T. Clearly, conditions (3.13.77) and (3.13.79) are satisfied
for (3.13.86).
The possible types of asymptotic behavior at infinity of nonoscillatory
solutions of (3.13.86) are as follows:
(I) limt ..... co x(t)/t = constant -=I 0,
(II) limt-+co x(t)/t = 0, limt-+oo Ix(t)1 =
(III) limt-+oo x(t) = constant -=I O.
00,
From Theorems 3.13.11 and 3.13.12 it follows that equation (3.13.86)

has a solution of the type (I) if and only if
00
si3q(s)ds <
(3.13.87)
00
r/'"
and that (3.13.86) has a solution of the type (III) if and only if
100 (100
q(u)du
ds <
00.
Theorem 3.13.13 implies that conditions (3.13.87) and
ioroo (100 q(u)du)

S
II",
ds =
00
(3.13.88)
are sufficient for the existence of nonoscillatory solutions y(t) of (3.13.86)

with A = 1 which satisfy limt-+oo ly(t)1 = 00.
Oscillation and nonoscillation of balf-linear differential equations
243
Conditions (3.13.87) and (3.13.88) are not always consistent. In fact,

let q(t) = (t+ l)m, where m is a constant. Then, (3.13.87) holds if and
only if m < -1 - {3, and (3.13.88) holds if and only if m 2: -1- a, and
hence these two conditions are inconsistent if a :S {3. Thus, if a > {3 and
- a :S m < -1 - {3, then there exists a solution of the type (II) for the
equation
(lx'(t)I" sgn x'(t))' + (t + l)Tnl x(t)l/3 sgn x(t) = O.

Next, we shall characterize types (I) and (III) solutions of a particular
case of (3.13.76), namely, the equation
(a(t)lx'(t)I,,-lX'(t))'
+ F(t, x[g(t)]) =
0,
a> 0
(3.13.89)
when the function F(t, x) is not necessarily nondecreasing in x.

Theorem 3.13.14. Let conditions (i) and (iii) (for (3.13.76)) hold,
(iv)' FE C([to, oo) x lR,lR) and sgn F(t, x)
= sgn x for
t 2: to.
Suppose for positive constants , and L with < L there exist positive
constants m and M depending on and L such that for :S Ixl :S L,
mF(t,):s IF(t,x)l:S MF(t,L),
t2:to.
Then equation (3.13.89) has a nonoscillatory solution

limHoo x(t) = constant -I- 0 if and only if
/00 a-l/,,(s) (1
00
IF(u, c)ldU) 1/" ds < 00
(3.13.90)
x(t)
such that
for some constant
-I- O.
(3.13.91)
Proof (The 'only if' part). Let x(t) be a nonoscillatory solution of
equation (3.13.89) such that limHoo x(t) = constant -I- O. Without loss of
generality, we assume that limt-too x(t) > 0, so that there exist positive
constants , Land tl 2: to such that :S x(t) :S Land :S x[g(t)] :S L
for t 2: tl. By (3.13.90), we have
F(t, x[g(t)]) 2: mF(t, )
(3.13.92)
for some constant m > 0 and t 2: tl.

Integrating (3.13.89) from s to t
we find for t 2: s 2: tl,
t 2: tb
it
F(u, x[g(u)])du
and noting that x'(t) > 0 for
a(s)(x'(s))" - a(t)(x'(t))" < a(s)(x'(s))",
which gives
a-l/,,(s)
(isroo F(u, x [g(u)])du )
1/0:
< x'(s),
Chapter 3
244
1:
a-I/<>(s)
(100
F(u, X[9(U)]dU) 1/<> ds < x(t) - x(t 1),
t 2 tl
which when combined with (3.13.92) yields
(The 'if' part). Suppose condition (3.13.91) holds for some constant c =f. O.
We can assume that c> O. By (3.13.90) there exists a constant M such
that c/2::::: x(t)::::: c implies F(t,x[g(t)])::::: MF(t,c) for t 2 t1 2 to
Choose T 2 t1 so large that T* = inft::::T get) and
M1/<>
00 a-1/<>(s)
(100
F(u, c)du
r/
<> ds <
and define the set Xc C([T*, (0)) and the mapping V: X --+ C([T*, (0))
by X = {x E C([T, (0)) : c/2 ::::: x(t) ::::: c, t 2 T*} and
Vx(t) =
-1
00
Vx(T)
a- 1/ <>(s)
for
(100
F(u, X[9(U)])dU) 1/<> ds
for
t2T
T*::::: t ::::: T.
Theorem 3.13.15. Let conditions (i), (iii) and (ivy hold, and suppose
that for positive constants f and L there exist positive constants m
and M (depending on f and L respectively) such that Izl::::: L implies
wet, zR[g(t), tlDI

and
Izl::::: f
:::
MF(t, LR[g(t), tIl),
2T
(3.13.93a)
IF(t, zR[g(t), t1DI,
2T
(3.13.93b)
implies
mF(t, fR[g(t) , hD
:::::
for any T 2 to large enough with inft>T g( t) 2 t 1 , where R[t, T] =

a-I/<>(s)ds and R(t) = R[t, to]. Then e~uation (3.13.89) has a nonoscillatory solution x(t) such that x(t)/R(t) --+ constant =f. 0 as t --+ 00 if
and only if for all large TI 2 T 2 to, g(t) 2 T for t 2 TI ,
f;
00
IF(s, cR[g(s), TDds <
00
for some constant
c =f. O.
(3.13.94)
245
Proof (The 'only if' part). Suppose equation (3.13.89) has a solution
x(t) such that limt~oo x(t)/ R(t) = constant i= O. We can assume that
limt~oo x(t)/R(t) > O. Then there exist positive constants i, tl and
T (T > t l ) such that x(t) ~ iR[t, tlJ and x[g(t)J ~ iR[g(t), tlJ for
t ~ T. The rest of the proof is similar to that of Theorems 3.13.11 and
3.13.13 and hence omitted.
(The 'if' part). Let condition (3.13.94) hold with c = 2k which can be assumed to be positive. By (3.13.93) there exists a constant !vI > 0 such that
o ~ z[g(t)J ~ 2k implies F(t, z[g(t)JR[g(t), tID ~ MF(t, 2kR[g(t) , tID,
get) ~ tl ~ to. Let T ~ tl be large enough and T* = inft>T get) ~ II
so that M
F(s,2kR[g(s),td)ds ~ (2'" -l)k"'. Consider the set Xc
C ([T* , ()()) and the mapping W: X --+ C ([T* ,
defined by
J;
(0
X = {x EC([T*,oo,
and
and
Wx(t)
a-l/",(s) [(2k)Q _
for
T*
for
kR[t,TJ~x(t)~2kR[t,tIJ
~ t ~
x(t) = 0 for
T*
t~T
T}
F(u,X[g(U)])dUf/ ds for t
~T
T.
Example 3.13.B. Consider the differential equation
(3.13.95)
0,
where m, n, J1, are positive constants, and II is any constant. The

function F(t, x) = t V lxl n - l x/(l + tJLlxl m ) satisfies conditions (3.13.90)
and (3.13.93), since 0 n.
It can easily be verified that (3.13.91) holds if and only if J1, > O! + II + 1
and (3.13.94) holds if and only if J1, + m > II + n + 1. Therefore, in view
of Theorems 3.13.14 and 1.13.15 necessary and sufficient conditions for
(3.13.95) to have nonoscillatory solutions x(t) satisfying limt~oo x(t) =
constant i= 0 and limHoo x( t) / R( t) = constant i= 0 are respectively,
J1, > O! + II + 1 and J1, + m > II + n + 1.
246
Chapter 3
Remark 3.13.2. If F(t, x) is nondecreasing, then conditioIls (3.13.90)

and (3.13.93) are trivially satisfied.
Remark 3.13.3. Suppose F(t, x) = q(t)f(x) where q(t) E ([to, (0), lR+),
f E C(ffi, JR) and sgn f(x) = sgn x. Then, F(t, x) satisfies (3.13.90).
If, in addition, f(x) has the property that klf(x)f(y)\:s If(xy)1
Klf(:r)f(y)l, xy 2:' 0 for some positive constants k and K, then
F(t, x) also satisfies condition (3.13.93).
:s
Next, we shall consider equation (3.13.89) when n(t) = ftCX) a-l/O(s)ds

satisfies n(to) < 00, to 2:' 0, and as before, we will classify all possible
nonoscillatory solutions according to their asymptotic behavior at infinity.
For this, we shall Ileed the following lemma.
Lemma 3.13.9. If x(t) is a nonoscillatory solution of equation (3.13.89)
which is either positive or negative on [to, (0), to 2:' 0 then there exist
positive constants CI and C2 such that
(3.13.96)
Proof. We can assume that x(t) > 0 for t 2:' to. In fact, a similar
argument holds if x(t) < 0 for t 2:' to. Since (a(t)lx'(t)IQ-Ix'(t))' =
-F(t, x[g(t)]) < 0 for t 2:' to, a(t)lx'(tlIQ-Ix'(t) is decreasing on [to, (0),
so that x' (t) is eventually of constant sign, i.e., either x' (t) > 0 for t 2:' to,
or there is some tl 2:' to such that x' (t) < 0 for t 2:' tl.
Suppose first that x'(t) > 0 for t 2:' to. Then, we have a(t)(x'(t))Q :S
a(to)(x'(to)a, t 2:' to which implies that x'(t):S al/a(to)x'(to)a-l/Q(t),
t 2:' to. Integrating this inequality from to to t, we find
x(t)
< x(to) + al/a(to)x'(to) (a-l/Q(s)ds
lto
< x(to) + a1/a(to)x'(to)n(to),
t 2:' to.
Suppose next that x'(t) < 0 for t 2:' tl. Then, since a(t)lx'(t)la-lx'(t)
= -a(t)(-x'(t)G, we have a(s)(-x'(sa 2:' a(t)(-x'(t)"', s 2:' t 2:' t l ,
or equivalently,
(3.13.97)
Dividing (3.13.97) by al/a(s) and integrating over [t, T],
x(t) > x(t) - X(T) 2:' - a1/CX(t)x'(t)

which in the limit as
--t
00
iT
we obtain
a-1/CX(s)ds,
T 2:' t
gives
x(t) 2:' - al/CX(t)x'(t)n(t),
2:' tl
(3.13.98)
247
Combining (3.13.98) with the inequality -al/<>(t)x'(t);::: -al/<>(tl)x'(tr),

t ;::: tl which follows from (3.13.97), we find x(t);::: -a l /a(t j )x'(t j )7r(t),
t ;::: tl. The conclusion (3.13.96) of the lemma now follows rather easily.
From Lemma 3.13.9 and its proof we can conclude that one and only
one of the following three possibilities occurs for the asymptotic behavior
of any nonoscillatory solution x(t) of (3.13.89).
= constant> 0,
(II) limt-+= x(t) = 0, liml.-->= l:r(t)I/7r(t) =
(III) limt-+= Ix(t)I/7r(t) = constant > o.
liml._,= Ix(t)1
(I)
00,
Now, we shall present sharp conditions which guarantee that (3.13.89)

possesses nonoscillatory solutions of these three types.
Theorem 3.13.16.
A necessary and sufficient condition for equation
(3.13.89) to have a nonoscillatory solution of the type (I) is that
= (
1.0
a(s)
for some constant c
1
s
to
IF(u, c)ldu
) l/a
ds <
00
(3.13.99)
# O.
Proof (i). (Necessity). Let x(t) be a solution of equation (3.13.89) of

the type (I). We may assume that x(t) > 0 for t ;::: to. A similar argument
holds if x(t) < 0 for t;::: to. As in Lemma 3.13.9, x'(t) is either positive
for t;::: to, or negative for t;::: tl for some tl ;::: to. Suppose x'(t) > 0
for t;::: to. An integration of (3.13.89) for t;::: to yields
1
t
to
F(s, x[g(s)])ds = a(to)(x'(t o))'" - a(t)(x'(t))a < a(to)(x'(to))CX,
which implies that Ito;;' F(s,x[g(s)])ds < 00. Combining this inequality
with the fact that x[g(t)];::: Cl, t ;::: tl ;::: to for some positive constant
Cl, we obtain Ito;;' F(s, cl)ds < 00. Hence, it follows that
1=(118
to
a(s)
1.0
F(u, cl)du
)1/0 ds:S;
7r(to)
Next, suppose that x'(t) < 0 for t >

(3.13.89) for t;::: tj gives
I.,
(1= F(s, cl)ds )l/C>

1.0
tl.
<
00.
Again an integration of
F(s,x[g(s)])ds = a(td(-x'(tl))c> - a(t)(-x'(t))c> < a(t)(-x'(t))C>,
248
Chapter 3
or
1
( a(t)
F( s, X[g( s )])ds
) 1/<>
:S - x' (t)
Integrating the above inequality over [tl' t2J and noting that C1 :S x[g( t) J :S
C2: t 2- iI for some positive constants C1 and C2, we find that
for any
t2
> t l , which implies (3.13.99).
(ii). (Sufficiency). \Vithout loss of generality we assume that the constant

c in (3.13.99) is positive. Choose T> to so large that T* = inft>T g(t) 2to and
-
(=
iT
(1a(s) iTt F(u, c)du)1/<> ds:S
(3.13.100)
and define the set X C C ([T* , (0)) and the mapping V : X ---+ C ([T* , (0))
by X = {x E C([T*, (0)) : c/2 :S :r(t) :S c, t 2- T*} and
c-l
Vx(t) = {
(a/s)
Vx(T),
lS
F(U,X[9(U)])dU) l/n ds,
t 2- T
T*:S t:S T.
In view of (3.13.100) it is clear that V maps X into itself. It can be verified

that V is continuous and V(X) is relatively compact in the topology
of C([T*,oo)). Therefore, the Schauder--Tychonov fixed point theorem
guarantees the existence of an element x E X such that x = V(x), i.e.,
x(t)
c-
iTt
(1a(s) iTt F(u, x[g(u)])du)
l/a
ds,
t 2- T.
Differentiation of this integral equation shows that x = x(t) is a positive

solution of (3.13.89) on [T, (0). It is clear that limt--+= x(t) = constant E
[cj2, c], which means that x(t) is a solution of the type (I). This completes
the proof.
Theorem 3.13.17.
A necessary and sufficient condition for equation
(3.13.89) to have a nonoscillatory solution of the type (III) is that
00
IF(s, C1T[g(S)]) Ids <
00
(3.13.101 )
to
for some constant c i=
o.
Proof (i). (Necessity). Let x(t) be a solution of the type (III) of

equation (3.13.89) which is eventually positive. Then there are positive
249

constants Cl, C2 and tl ~ to such that Cl7r[g(t)] ~
for t ~ h. We can suppose that x'(t) < 0 for t ~
(3.13.89), we have a(t)( -x'(t))Q ~
F(s, x[g(s)])d.s,
this with the inequality a(t)(-x'(t))Q ~ (x(t}/7r(t))'"
is equivalent to (3.13.98), we obtain
Itt,
It
It
F(.s, Cl7r[g(s)])d.s
F(s, x [g(s)])d.s
Thus, (3.3.101) follows in the limit as t -+
~ (;~~D
x[g(t)] ~ C27r[g(t)]
h. From equation
t ~ tl. Combining
for t ~ to, which
<
c~
for t
~ t2
00.
(ii). (Sufficiency).
We only need to consider the case when the constant C in (3.13.101) is positive. Let k > 0 be a constant such that
2k < cQ and take T > to so large that T* = inft2:T get) ~ to, and
F(s, (2k)1/Q 7r [g(.s)])ds ~ k. Consider the set
I;
Z = {z E C([T*, (0)) :
k l / Q7r(t) ~ z(t) ~ (2k)1/Q 7r (t), t ~ T,

kl/Q7r (T) ~ z(t) ~ (2k)1/Q 7r (t), T* ~ t ~ T}
and the mapping
Wz(t) = { lex)
(a~) (k + irs F(U,Z[9(U)]dU)) 1/0 d.s,
Wz(T),
If z E Z,
then for t
kl/o lex)a-l/Q(.s)d.s
T*
~ t ~
t ~T
T.
(3.13.102)
T, we have
~ Wz(t)
(t
\
k + iT F(u, (2k/ IQ 7r[g(u)]du)
it(ex)a-l/Q(s)
(2k)1/Q lex) a-1/Q(.s)d.s
l/Q
ds
= (2k)1/Q 7r (t),
which implies that W z E Z. Therefore, W maps Z into itself. Since the

continuity of Wand the relative compactness of W(Z) can be proved
without any difficulty, there exists an element z E Z such that z = W z,
i.e.,
From this integral equation it follows that z(t) satisfies equation (3.13.89)
on [T, (0), and by L'Hospital's rule limHex) z(t)/7r(t) = constant E
Chapter 3
250
[kI/e> , (2k)I/e>]. Thus, z(t) is a solution of the type (III) of (3.13.89). This
Theorem 3.13.18. The equation (3.13.89) has a solution of the type (II)
if
8
)1/e>
(3.13.103)
100(11
a(s)
to
to
ds <
IF(u, c)ldu
00
for some constant c i- 0, and
100
IF(s,d7r(s))lds
(3.13.104)
00
to
for any constant d with cd> O.

Proof. We shall only consider the case where c> 0 in (3.13.103). Choose
k > 0 so that (2k)1/e> [1 + 7r(to)] :::; c, and let T> to be large enough
such that T* = inft~T get) ~ to and
1 r
iTroo ( a(s)
iT F(u, c)du) 1/e> ds:::;
k 1/e>.
Then it can be verified by the Schauder fixed point theorem that the mapping W defined by (3.13.102) has a fixed element in the set
Y
= {y
E C([T*,
00)) :
k 1/ e>7r(t):::; y(t):::; c, t ~ T
k 1/ e>7r(T) :::; yet) :::; c, T*:::; t :::; T}.
In fact, if y E Y, then
Wy(t)
<
100
a-1/n(s) [(2k)1/e>
< (2k)1/e>7r(t) + 21/e>
100
(2 s F(U,Y[9(U)])dU) lie>] ds
(a(s)
< (2k)1/e>[1 +7r(to)] :::; c,
8 F(u, Y[9(U)])dU) lie> ds
t~T
and so W y E Y. The verification of the continuity of Wand the relative

compactness of W (Y) are as before. Therefore, there exists a fixed element
yET which is a solution yet) of equation (3.13.89) for t ~ T*. It is
clear that yet) --+ 0 as t --+ 00. Further, in view of (3.13.104), l'Hospital's
rule gives
'et)
lim '1L-() =
7r ' t
lim
t ...... oo
t ...... oo
>
lim (k
t ...... oo
lim
t ...... oo
iTrt F(s, k
(jt
)1/0:
k+
F(s,y[g(s)])ds
T
1/0:7r(S)) ds
00.

This shows that y(t) is a type (II) solution of (3.13.89).
251
(3.13.105)
where a, /3, A are positive constants and J.l is any constant. Applying
Theorems 3.13.16 and 3.13.17 to (3.13.105), respectively, we find
(i) equation (3.13.105) possesses a nonoscillatory solution of the type (I)
(which is bounded from below and above by positive or negative constants)
if and only if J.l < A, and
(ii) equation (3.13.105) possesses a nonoscillatory solution of the type (III)
(which behaves like a constant and multiple of e-(>./a)t as t -+ 00) if and
only if J.lIA < /3/a.
Theorem 3.13.18 applies only to the case where a > /3. In fact, in this
case the condition /31 a :':::: J.lI A < 1 ensures the existence of a nonoscillatory
solution x(t) of equation (3.13.105) satisfying limHoo x(t) = 0 and
limHoo e(>.}<tlx(t) I = 00.
3.13.5. A Comparison Theorem

Most of the known oscillation results for functional differential equations
are generalizations of those existing for the ordinary differential equations.
Further, often the method of the proof of a generalized result is the same
as that of the original result, and hence imposes a severe restriction on the
deviating argument. Here, we shall reduce such a generalized problem to
an ordinary differential equation so that the oscillation criteria presented
in Sections 3.8 - 3.11 can be employed directly.
We need the following lemma.
Lemma 3.13.10. Suppose g(t) E C([to, oo),JR), FE C([to, 00) x JR, JR),
g(t) :':::: t for t 2>: to, limHoo g(t) = 00, xF(t, x) > 0 if x -# 0, t?: to and
F(t, x) is nondecreasing in x. If the differential inequality
y'(t) - F(t, y[g(t)]) ?: 0
(3.13.106)
has a positive solution on [tl' 00) for some tl ?: to, so does the equation
y'(t) - F(t, y[g(t)]) = O.
(3.13.107)
Proof. Let y(t) be a positive solution of (3.13.106) on T?: tl so

that g(t) ?: tl for t?: T. Then, y(t) satisfies the inequality y(t) ?:
Chapter 3
252
y(T)
+ J; F(u., y[g(u)])du.
zl(T)
zn(t)
Let
y(t) for t 2 t l ,
y(t) for t E [tl,T]
ht
F(u, Yn-l[g(u)])du for t 2 T, n = 2,3, ....

(3.13.108)
It follows from the definition of Zn and (3.13.108) that the sequence {Zn}
satisfies y(t) = zdt) 2 Z2(t) 2 ... 2 y(T) for all t 2 T. Hence, {zn}
converges pointwise to a function z(t), where y(t) 2 z(t) 2 y(T) for
all t 2 T. Let Fn(t) = F(t, Z,,[g(t)]) , n = 1,2,. Then, FI (t) 2
F 2 (t) 2 ... 2 O. Since FI is integrable on [T, t] for any t 2 T,
and limn--+oo Fn(u) = F(u, z[g(u)]) for any u E [T, t], by the monotone
convergence theorem, we have z(t) = y(T)+ J;F(u, z[g(u)])du for t 2 T.
Hence, z(t) satisfies equation (3.13.107).
y(T)
Now, we shall prove the following main result.

Theorem 3.13.19.
Let the functions
and
be as in Lemma
3.13.10 and suppose g(t) E CI([to, 00), JR.) and g'(t) > 0 for t 2 to 20,
a(t) E C([to, oo),JR.+) and Joo a-l/O(s)ds = 00, and a> 0 is a constant.
If (3.13.89) has a nonoscillatory solution, then the equation
(a(s)ly'(s)IO-Iy'(s))'
+ g'[g2 1 (S)]F(9-I(s),y(s)) =
0,
('
= :s)
(3.13.109)
has a nonoscillatory solution.

Proof. Let x(t) be a llonoscillatory solution of equation (3.13.89) and
assume that x(t) > 0 for t 2 to 2 O. Clearly, there exists a tl 2 to such
that x'(t) > 0 for t 2 tl. Integrating (3.13.89) from s to T, T 2 s 2 t l ,
we obtain
a(T)(x'(T))o. - a(s)(x'(s))O
iT
F(u, x [g(u)J)du = 0
and hence x'(s) 2 a-l/a(s)(Jsoo F(u,x[g(u)])du)l/a, s 2 tl' Let v

g( u), then this inequality yields
x'(s) 2 a-l/o(s) (
roo F(g~l~;, x(v)) dV) 1/0,

Jg(s)
9 (g
(v))
s 2 tl'
Since g(t):S t, we get
x'(s) 2 a-1/a(s) (
roo F(g-l(v), x(v)) dV)
Js
g'(g-l(v))
1/0
Tx(s),
s 2 iI
253
By Lemma 3.13.10, the equation x'(s) = Tx(s) has a positive solution

yes) on [tl' (0) such that y'(s) > 0 for all s 2': t l . Differentiating the equation y' (s) = Ty( s), we find that y( s) satisfies equation
(3.13.109).
By Theorem 3.13.19 it follows that the retarded half-linear equation
(lx'(t)I"-lX'(t))' + q(t)lx[g(t)]I"-lx[g(t)]
0,
(3.13.110)
where q(t) E C([t o, (0), lR+), get) E C([to, (0), lR), get) :S t, g'(t) > 0 for
t 2': to 2': 0, limt-+oo g( t) = 00, and a > 0 is a constant, is oscillatory if
the ordinary half-linear equation
(ly'(s)IO-ly'(s))' +
(:}~~ll~~]]) ly(s)IO-ly(s)
= 0,
(' =
:s)
(3.13.111)
is oscillatory. Further, applying Corollary 3.13.3, we find that if
oo
,\ q[g-l(S)]
s g'[g_l(s)]ds =
00,
(3.13.112)
0 <). < a
then equation (3.13.111) is oscillatory. If we let s = get), then (3.13.112)

reduces to
JOO g'\(t)q(t)dt
(3.13.113)
00.
Thus, (3.13.113) implies that (3.13.111) and hence equation (3.13.110) is

oscillatory.

Consider equation (3.13.2) subject to the conditions (i) - (iv), and let
x(t) be its nonoscillatory solution. Clearly, x'(t) is eventually of constant

sign, so that
either
x(t)x'(t) < 0
or
x(t)x'(t) > 0 for
2': T
(3.13.114)
provided T is sufficiently large. Thus, x(t) is bounded or unbounded

according to whether the first or the second inequality in (3.13.114) holds.
Our first result shows that in the case when get) is a retarded argument,
equation (3.13.2) may not have any unbounded nonoscillatory solution.
Theorem 3.13.20. Suppose get) < t for t 2': to. Then every bounded
solution of equation (3.13.2) is oscillatory if either one of the following
conditions holds
(C l
limsup
q(s)R"'[g(t),g(s)]ds> 1,
t--+oo } g(t)
(3.13.115)
Chapter 3
254
(C 2 )
lim sup
t-->oo
(C 3 )
it
a-l/",(s)
g( t)
(it
q(U)dU) II", ds > 1,
(3.13.116)
every solution of the delay equation
y'(t)
+ q(t)R"'[(t + g(t))/2,g(t)]y[(t + g(t))/2]
(3.13.117)
q(s)RO[(s + g(s))/2,g(s)]ds > 1
(3.13.118)
is oscillatory, or
(C 4 )
liminf
t-->oo
(t
i(t+g(t/2
Proof. Let x(t) be a bounded nonoscillatory solution of equation (3.13.2).

Without loss of generality we can assume that x(t) is eventually positive.
Thus, there is a To:2: to such that x( t) > 0 and x' (t) < 0 for t:2: To.
Suppose (3.13.115) holds. Let T > To be such that inft>T get) :2: To.
Applying Lemma 3.13.2 with g(s) and get) for CY and T respectively, in
the inequality (3.13.11), we get x[g(s)]:2: _al/o [g(t)]R[g(t), g(s )]x'[g(t)]
for t:2: s :2: T, which implies
q(s)x'" [g(s)] :2: q(s)R"'[g(t),g(s)](a[g(t)](-x'[g(t)])O,
t:2: s:2: T.
Replace the left-hand side of the above inequality by (a(s)lx'(s)IO-Ix'(s))'

and integrate from get) to t, to obtain
= - (a(s)(-x'(s))"')'
a[g(t)]( -x'[g(t)])O- a(t)(-x'(t))'" :2: a[g(t)]( -x' [g(t)])'"

x
it
q(s)R<>[g(t),g(s)]ds, t:2: T
get)
and hence
a[g(t)](-x'[g(t)])O
[it
q(s)RO[g(t),g(s)]ds
get)
-1] : :
t:2: T.
for
But this is inconsistent with (3.13.115).

Suppose now that (3.13.116) holds. Integration of equation (3.13.2) over
[CY, t] gives
a(t)( -x'(t))O +
a(cy)( -x'(cy))"
>
it
it
q(U)x'" [g(u)]du
q(U)X"'[g(u)]du for t:2: CY
~ T,
which implies
-x'(CY) :2: a- 1/ o (cy)
(it q(U)x'" [g(U)]dU) 11"',
t:2: cy:2: T.
(3.13.119)
255
From (3.13.119) and the equality
x(s) = x(t)
we find
.r
x(s) 2:
Putting s
we get
x[g(t)]
it
a-l/"'(u)
(-x'(u))du
for
t 2: s 2: T,
(1 q(U)x'" [9(U)]dU)
t
1/",
(3.13.120)
t 2: s 2: T.
du,
g(t) in (3.13.120) and using the fact that x[g(t)] is decreasing,
)
l(t) a-1/",(u) (1t
q(u)d'U
t
l/n
1<
du - 1
for
t 2: T,

T
Next, we apply Lemma 3.13.2 with g(t) and (t

respectively, in inequality (3.13.11), to get
x[g(t)] 2: a-lin
w(t)
Using this inequality in (3.13.2) and letting
+ q(t)R'"
[g(t), t
+ ~(t)]
CXJ
11
q(s)R'" [g(s), s +~(s)] w [s
t 2: T.
a(t)( -x'(t))'"
[t +~(t)] ::;
Integrating the above inequality from t to

find
w(t) 2:
for u and
[t +tt)] R [g(t), t +~(t)] (-x' [t +~(t)]),
t 2: T, we obtain
w'(t)
+ g(t))/2
for
0 for t 2: to.
and letting 'U --+
+~(s)]
ds,
00,
we
t 2: T.
But, then in view of Lemma 3.13.6, equation (3.13.117) has a positive

solution y(t) with limHCXJ y(t) = 0, which is a contradiction.
Finally, we note that condition (3.13.118) ensures the oscillation of equation (3.13.117) (d. Lemma 3.13.4 (a3)). This completes the proof.
Example 3.13.10. The equation
(e-tlx'(t)lx'(t))'
Ix[t/2]lx[t/2]
(3.13.121)
has an unbounded nonoscillatory solution x(t) = et . The hypothesis of

Theorem 3.13.20 (C l ) is satisfied and hence all bounded solutions of equation (3.13.121) are oscillatory.
256
Chapter 3
A duality of Theorem 3.13.20 holds in the case when get) is an advanced

argument.
Theorem 3.13.21. Suppose get) > t for t 2 to. Then every unbounded
solution of equation (3.13.2) is oscillatory if either one of the following
conditions holds
(AI)
(A 2 )
(A3)
i 9 q(s)R"[g(s),g(t)]ds> 1,
i 9(t) (a(s)1 i q(u)du 1/" ds> 1, or
lim sup
i(t+9(t))/2 q(s)R" [g(s),
s + g(s)]
1
liminf
ds> -.
lim sup
(t)
t ..... oo
t ..... oo
t ..... oo
(3.13.122)
(3.13.123)
(3.13.124)
Proof. Let x(t) be an unbounded nonoscillatory solution of equation

(3.13.2) which can be assumed to be eventually positive. There is a To 2 to
such that x(t) > 0 and x'(t) > 0 for t 2 To.
Suppose (3.13.122) holds. Letting {}" = g(s) and T = get) in the

inequality (3.13.12), we get x[g(s)] 2 R[g(s), g(t)]a 1 /" [g(t)]x'[g(t)], s 2
t 2 T where T is as in Theorem 3.13.20. Thus, for s 2 t 2 T,
(a(s)(x'(s))")' = q(s)x"[g(s)] 2 q(s)R"[g(s),g(t)]a[g(t)](x'[g(t)])".

Integration of the above inequality over [t, get)] yields
[i
a[g(t)] (x' [g (t)]) "
9 (t)
1<
q(s)R"[g(s),g(t)]ds - 1
0,
which is a contradiction to (3.13.122).

Suppose (3.13.123) holds. Combining
x'({}") 2 a- 1 /"({}")
with the relation xes)
xes) 2
(i
t
er
q(u)x"[g(u)]du
)1/" ,
= x(t) + J/ x'({}")d{}", s 2 t 2 T
1
a({}")
er
q(u)x"[g(u)]dU ) 1/" d{}",
we obtain
s 2 t 2 T.
Putting s = get) and noting that x[g(t)] is increasing, we obtain
x[g(t)]
[itret)
(1 itr
a({}")
q(u)du
)1/" d{}" -11 ~
0 for
2 T,
257

Suppose (3.13.124) holds. Letting ()"
inequality (3.13.12), we get
x[g(t)]
aI/a [t
= g(t) and
= (t + g(t))/2 in
+:(t)] R [g(t), t +:(t)] x' [t +:(t)] ,
Using this inequality in (3.13.2) and letting y(t)

obtain
y'(t)
q(t)RO [g(t), t
= a(t)(x'(t))O, t
~ T.
~
T we
+:(t)] y [t +:(t)] , t ~ T.
But, in view of Lemma 3.13.5 and condition (3.13.124), the above inequality has no eventually positive solutions, which is a contradiction. This
Example 3.13.11. The equation
(e-tlx'(t)lx'(t))'
3et lx[2t]lx[2t],
t~0
(3.13.125)
has a bounded nonoscillatory solution x(t) = e- t . Clearly, the condition

of Theorem 3.13.21(A l ) is satisfied and hence all unbounded solutions of
(3.13.125) are oscillatory.
Remark 3.13.4. Theorems 3.3.20 and 3.3.21 also hold for the differential
inequalities of the form
{(a(t)lx'(t)IO-lx'(t))' - q(t)lx[g(t)]IO-lx[g(t)]} sgn x[g(t)] ~
o.
(3.3.126)
Remark 3.13.5. Theorems 3.13.20 and 3.13.21 fail to apply to equations
of the type (3.13.2) and/or inequalities of the form (3.13.126) when g(t) =
t. Also, oscillation of all solutions of (3.13.2) does not follow from Theorem
3.13.20 or 3.13.21, because the deviating argument under consideration is
only either retarded or advanced. However, for equations of mixed type, i.e.,
involving both retarded and advanced arguments, it is possible to establish
oscillation of all solutions. We shall show this for the mixed type equation
n
(a(t) lx' (t) la-lx, (t))'
L qi(t) IX[gi(t)]10-
X[gi( t)],
(3.13.127)
i=l
where
(i)
(ii)
>0
is a constant,
qi(t) E C([to, 00), JR.+) such that
to, i = 1,2, ... ,n,
a
SUp{qi(t); t ~ T} > 0 for any
Cha.pter 3
258
(iii) gi(t) E CI([to, 00), JR), g~(t) 2: 0 for t 2: to and
00, i = 1,2"", n,
(iv)
a(t)
C([to, 00), JR+) and
limt-+oo gi(t) =
J'X) a-I/Q(s)ds = 00.
The following theorem follows from the above results.

Theorem 3.13.22 (MI)' All bounded solutions of equation (3.13.127) are
oscillatory if there exists an i E {I, 2, ... , n} such that gi (t) < t for
(C 4 ) of Theorem 3.13.20
t 2: to, and either one of the condition~ (Cd
holds with 9 and q replaced by gi and qi respectively.
(M 2) All unbounded solutions of equation (3.13.127) are oscillatory if there

exists a jE{1,2,,n} such that gj(tt for t2:to, and either one
of the conditions (AI) - (A3) of Theorem 3.13.21 holds with 9 and q
replaced by gj and qj respectively.
(M3) All solutions of equation (3.13.127) are oscillatory if there exist i

and j E {I, 2"", n} such that gi(t) and gj(t) satisfy the conditions
(M I ) and (M2) respectively.
Proof (MI)' Suppose to the contrary that equation (3.13.127) has a
bounded nonoscillatory solution x(t). Then from (3.13.127) it follows that
x(t) satisfies the differential inequality
{(a(t)lx'(t)la-Ix'(t))' - qi(t)lx[9i(t)]ln-Ix[9i(t)]} sgn X[gi(t)] 2: 0

(3.13.128)
for all sufficiently large t. This is however impossible, because the existence
of a bounded nonoscillatory solution of (3.13.128) is excluded by Theorem
3.13.20 and Remark 3.13.4.
(]'vIz) An unbounded nonoscillatory solution x(t) of equation (3.13.127),

if exists, satisfies the inequality (3.13.128) withi replaced by j for
sufficiently large t. But this is impossible because of Theorem 3.13.21 and
Remark 3.13.4, and hence every unbounded solution of (3.13.127) must be
oscillatory.
(l\Ih) This is an immediate consequence of (l'vh) and (M2).
Example 3.13.12. For the equations
klx[t - u]la-lx[t - u],
(3.13.129)
(lx'(t)la-IX'(t))' = Clx[t + T]la-Ix[t + T],
(3.13.130)
(Ix' (t)
(Ix' (t) la-Ix' (t))'
1 x'
(t))'
= klx[t-u] Ia-Ix[t-u]+Clx[HT]la-lx[HT], (3.13.131)
where a, k, C, u and
are positive constants, we find
259
(1)
from conditions (C I ) or (C 2 ) of (M I ) of Theorem 3.13.22 that all
bounded solutions of (3.13.129) are oscillatory if
a+1
a
> --,
(3.13.132)
(2)
from conditions (Ad or (A2) of (M2) of Theorem 3.13.21 that all
unbounded solutions of (3.13.130) are oscillatory if
70+1
> a + 1 or
j71/"T(o+I)/0
> a + 1.
a
'
(3.13.133)
(3)
from the last statement of Theorem 3.13.22 tha.t all solutions of
(3.13.131) are oscillatory provided both (3.13.132) and (3.13.133) hold.
3.13.7. Nonoscillation of Solutions of

Equation (3.13.127)
We shall now study the existence and asymptotic behavior of nonoscillatory solutions of equation (3.13.127). If x( t) is a nonoscillatory solution
of (3.13.127), then there exists a tl > to such that either
x(t)x'(t) > 0
for
t:::., tl,
(3.13.134)
or
x(t)x'(t) < 0 for t:::., tl'

(3.13.135)
If (3.13.134) holds, then x(t) is unbounded and the limit a1/Q(00)x'(00) =
limt"'HXJ a l /<:> (t)x'(t) exists and is finite or infinite, and if (3.13.135) holds,
then x(t) is bounded and the finite limit x(oo) = lirnHCXl J;(t) exists.
In what follows we only need to consider eventually positive solutions
of (3.13.127), since if x(t) satisfies (3.13.127), then so does - x(t). Let
x(t) be an eventually positive solution of (3.13.127) satisfying (3.13.134)
and having a finite limit al/Q(oo)x'(oo) = limHoo a1/Q(t)x'(t) > O. Then
twice integration of (3.13.127) for t:::., T yields
x(t)
~ .x(T) I J:a -, 1"(,) [a( 00 )(.x' (00))"
-1= ~q,(
a).x" Iq,( a)lda 'I" d.,

(3.13.136)
where T> tl is chosen so that inft:2:T 9i(t) :::., tl, i
= 1,2"", n.
Similarly, if x(t) is an eventually positive solution of (3.13.127) satisfying (3.13.135), then integration of (3.13.127) twice from t to v and the
letting v 4 00 gives
x(t)
(3.13.137)
Chapter 3
260
Based on these integral representations (3.13.136) and (3.13.137) of

(3.13.127), we shall prove the following existence theorems.
Theorem 3.13.23. The equation (3.13.127) has a nonoscillatory solution
x(t) such that limHoox(t)/R(t) = constant =1= 0 if and only if for all
large T,
./00 qi(s)RD.[gi(S), T]ds
< 00,
i = 1,2,, n.
(3.13.138)
Proof (The 'only if' part). Let x(t) be a nonoscillatory solution of

equation (3.13.127) satisfying limHoo x(t)/ R(t) = c > 0, c is a constant.
./00
qi (s)xD. [gi(s)]ds < 00.
i=l
This combined with the relation limHoo X[gi(t)l! R[gi(t), to]

1,2,, n where T 2: to is large so that inft>T gi(t) 2: to
(3.13.138).
-
c, i
implies
(The 'if' part). Suppose (3.13.138) holds. Let k > 0 be arbitrarily fixed,
and let T > to so large that
(3.13.139)
and
Consider the set X

defined by
C ([T , 00)) and the mapping W: X -+ C ([T , 00))
X = {x E C([T*, 00)) : (k/2)R[t,T]:S x(t):S kR[t,T], t 2: T,
x(t)
0, T*:S t :S T}
and
Wx(t)
h
t
a- 1 /"(s) k"
-1
00
t;qi(U)X"[9i(U)]dU
]1/0.
ds, t 2:: T
0, T*:S t :S T.
It is clear that X is a closed convex subset of the Fnchet space C ([T, 00))
of continuous functions on [T, 00) with the usual metric topology, and
261
W is well-defined and continuous on X. It can be shown without difficulty that W maps X into itself and W(X) is relatively compact in
C( [T*, (0)). Therefore, by the Schauder-Tychonov fixed point theorem,
W has a fixed element x in X which satisfies
x(t)
h
t
a- 1/ c>(s) kQ
-.i 8
00
q;(u)xc>[g;(u)]du
ll/Q
ds,
t '2 T.
Finally, twice differentiation shows that x( t) satisfies eqnation (3.13.137)

for t '2 T and limt-+oo x(t)/ R(t) = lil1lHoo a 1/ e>(t)x' (t) = k.
Theorem 3.13.24. The equation (3.13.127) has a nonoscillatory solution

x(t) such that limHoo x(t) = constant f 0 if and only if
J [1
DO
a(s)
00
q;(u)du ]
1/0
ds <
00,
i=1,2,,n.
(3.13.140)
Proof. The 'only if' part follows from (3.13.137). To prove the 'if' part,
snppose (3.13.140) is satisfied. Choose T> to so that (3.13.39) holds and
]lie> ds::S:
iT(exo [a-1/e>(s) 100
s
q;(u)du
and define X
X
c C ([T* , 00 ))
= {x
2"
and V: X -+ C ([T* , 00 ) ) by
E C([T*,
(0)) : k::S: x(t)::s: 2k, t '2 T*},
where k > 0 is a fixed constant, and
Vx(t)
kl
1=
Vx(T),
a-'I"(.,)
[1.= ~q'(U)X"l.q'(U)ldUr" d.,.
t ';> T
T*::S: t ::s: T.
As in Theorem 3.13.23 one can verify that V maps X into a relatively

compact subset of X so that there exists a x E X such that
Finally, twice differentiation shows that x(t) satisfies equation (3.13.127)

for t '2 T and x(t) -+ k as t -+ 00.
It remains to discuss the existence of an unbounded nonoscillatory solution x(t) of (3.13.127) which has the property that limHexo x(t)/ R(t) =
00,
and of bounded solution x(t) of (3.13.127) having the property that
Chapter 3
262
limt-+oo x(t) = O. However, this is a difficult problem and there is no general criteria available for the existence of such solutions. Therefore, we
confine ourself to the case where at least one of the gi(t) is retarded and
present sufficient conditions which guarantee that (3.13.127) has a nonoscillatory solution which tends to zero as t -+ 00. Such a solution is often
referred to as a decaying nonoscillatory solution. Our derivation is based
on the following theorem.
Theorem 3.13.25. Suppose

io E {I, 2, ... , n} such that
00
a- 1 /<>(s)ds
00,
and there exists an

(3.13.141)
Further, suppose there exists a positive decreasing function yet) on [T,oo)

satisfying
yet) ~
1
t
00
a- 1 /<>(s)
Jsroo qi(u)x<>[gi(u)]du]1/<> ds,
t ~T
(3.13.142)
where T is chosen so that inft>T gi(t) ~ to, i = 1,2, ... ,n. Then equation
(3.13.127) has a nonoscillatory -solution which tends to zero as t -+ 00.
Proof. Let Z = {z E C([T,oo: 0 < z S; yet), t ~ T}. With each
E C([to, 00 defined by
z E Z we associate the function
z(t)
= {
z(t)
z(T)
for
>T
+ [yet) -
yeT)]
Define the mapping H: Z -+ C([T, 00
for
(3.13.143)
to S; t S; T.
as follows
t ~ T.
Clearly, H is continuous and maps Z into a relatively compact subset

of Z. Therefore, it follows that there exists a z E Z such that z = Hz,
i.e.,
~T.
(3.13.144)
Twice differentiation of (3.13.144) gives
n
(-a(t)(-z'(t<'
Lqi(t)(Z[gi(t)j)",
i=l
t ~T
Oscillation and non oscillation of balf-linear differential equations
263
which in view of (3.13.143) implies that z(t) is a solution of (3.13.127)

for all sufficiently large t. It remains to show that z(t) is positive. For
this, we observe that z(t) is positive on [to, T] and to:; g;o(T) < T
for some io E {1,2,,n}. Therefore, it follows that z[g;o(T)] > 0 for
some io E {l,2,,n}, and hence L~=lq;(T)(z[g,(T)])a > O. Thus,
(-a(t)( -z'(t))Q)'lt=T f o. Furthermore, we find that z is nonnegative
and decreasing on [T, (0). Hence, if z(T) = 0, then z is icienticCl]]Y zero
on the whole interval [T, (0), and therefore (-a(t)( -z'(t))c'lt=T = 0,
which is a contradiction. Thus, we have shown that z(T) > O.
Next, let T* be the first zero of z in (T, (0). Then, z is positive on
[to, T*), and hence L:'=l qi (T*)(z[g; (T)J)<' > 0, since to :; .% (T*) < T*
for some io E {I, 2"", n}. Hence, (-a(t)( -z'(t))c'lt=To f O. On
the other hand, z is identically zero on [T* ,(0), which means that
(-a(t)( -z'(t))Q)'lt=T* = 0, a contradiction. Therefore, the solution z(t)
is positive. This completes the proof.
To apply Theorem 3.13.25 in practice, we need to distinguish the following three cases:
JOO
q;(s)ds <
00
<
and
i=l
JOO
tlJ
;=1
and
00,
(3.13.145)
i (8)ds
<
00
and
./~ "-'1"(.,) [1= t,q'(U)dUr" d.,
oc
(3.13.146)
J=L
n
q;(s)ds =
00.
(3.13.147)
;=1
The condition (3.13.145) which is the same as (3.13.140) guarantees the

existence of a decaying nonoscillatory solution of equation (3.13.127).
Theorem 3.13.26. Suppose (3.13.141) holds for some io E {I, 2,,, n}.
If condition (3.13.140) is satisfied, then equation (3.13.127) has a nonoscillatory solution which tends to zero as t --t 00.
Proof. Let T be large enough so that min;{inft~T g;(t)} ~ max{l, to}
and
(3.13.148)
Chapter 3
264
We shall show that y(t) = 1 + (lit) satisfies (3.13.142). Indeed, we have
The conclusion now follows from Theorem 3.13.25.
Now, we shall present results which guarantee the existence of decaying nonoscillation solutions and are applicable for the cases (3.13.146) and
(3.13.147).
Theorem 3.13.27.
n}, and
lim sup
t-+oo
00
Suppose (3.13.141) holds for some io E {I, 2,,
[1 t
aCt)
~,
Qi(U)dU] 1/0< ds <
00
a-I/,,(s)
i=I
(3.13.149)
where a(t) = mini gi(t). Then equation (3.13.127) has a nonoscillatory

solution which tends to zero as t -+ 00.
Proof. Let
Q(t)
a~t)
1 ~Qi(S)dS
00
1/"
and choose T> to so that inft2::T a(t) 2:: to, and
QT = sup
t2::T
Define y(t)
= exp ( -(l/QT)
y[gi(t)]
exp (Q1
It
Q(s)ds <
aCt)
It: Q(s)dS).
t
ThiW
< e exp ( - dT
(3.13.150)
Since for i
Q(S)dS) exp
1:
= 1,2,, n,
(_QlThort Q(S)dS)
Q(S)dS) = ey(t),
t 2:: T
265
in view of (3.13.150), we have
r( t t,
at,)
< e
1=
q, (<<)(ylq, (u)J)"
Q(s)y(s)ds :.:; e
< eQTexp (- dT
1:
00
dU) 'In d.,
Q(s) exp ( - dT
1:
Q(u)dU) ds
= eQTy(t) :::; y(t),
Q(S)dS)
t:::: T.
Now, from Theorem 3.13.25 it follows that equation (3.13.127) has a decaying nonoscillatory solution.
Theorem 3.13.28. Suppose (3.13.141) holds for someio E {I, 2"", n}.
Further, suppose there exists aT> to such that inft2'T a(t) :::: to,
r=
inf Q(t) > 0 and sup

t2'T
t2'T.
O'(t)
(Q* )<>/(0+1)
+1
Lq;(u)du < _a_

i=l
e
--.I..
(3.13.151)
where the functions Q(t) and a(t) are as in Theorem 3.13.27. Then
equation (3.13.127) has a nonoscillatory solution which tends to zero as
t ---+
00.
Proof. Let
PT = sup
t2'T
it
and
y(t) =
exp (-
(~)
aPT
Then, we have y[gi(t)] :::; exp ((a

hence
1 ~qi(S)(Y[9i(S)WdS
1
00
e<>+l
00
<
a
1t
tq;(S)dS) .
t Oi =l
+ l)ja) y(t),
t :::: T, i = 1,2,,,,, nand
1= (~q;(S))
+ 1
---p:-
:::; e<>+l
Ln q;(s) exp
i=l
PT e<>+l exp (- a p+
+1
q;(t)ds
<T(t) ;=1
11t
to ;=1
(y(s))<>ds
qi(u)du ) ds
to
q;(S)dS) '
t:::: T.
266
Chapter 3
r ( rt,
Consequently, we obtain
at,)
_P_
T_ ) l/n
~ (
+1
QT
~
~
e(a+1)/<>
__
T_
a 1
l/n
e(o+l)/a
for
exp
100
d,
_a__1
aPT
qi(S) exp
1/0. e(a+l)/a
L qi(u)du
n
ds
(+ 118 L
+ I1t t
to i=l
aPT
(PT )
(a+l)QT a+l
yet)
100
t
1(P,+
~ -*
r
(+ 18
q,(.)(y 19, (u)l)"d.
__
a__
to i=l
q;(S)dS)
exp (_ a
aPT
qi(u)du ) ds
aPT
tO i =l
T.
This establishes the existence of a strictly decreasing function yet)

satisfying (3.13.142). The proof now follows from Theorem 3.13.25.
>0
(lx'(t)lx'(t))' = c'Ylx[Atllx[At],
,>
(3.13.152)
t> 0
where
1, 0 < A < 1 are constants. This is a special case of (3.13.127)
with a = 2, aCt) = 1, n = 1, ql(t) = r'Y and gl(t) = At.
(11) Let, > 3. Then both (3.13.138) and (3.13.140) hold for (3.13.152)
and so by Theorems 3.13.23 and 3.13.24, equation (3.13.152) has nonoscillatory solutions X1(t) and X2(t) such that limHoo Xl(t)/t = constant of. 0
and limt-+oo X2(t) = constant of. 0 regardless of the values of A, 0 < A < 1.
(12) Let, = 3. An easy computation shows that (3.13.149) is satisfied
for (3.13.152) if 1 < A < exp(vI2/e), since
1
t
g,(t)
(100 ql (u)du ) 1/0. ds

8
it (1
At
00
u- 3 du )
1/2
ds
>:.
1In In 1
v2
From Theorem 3.13.27 it follows that for each A equation (3.13.152) has
a nonoscillatory solution which tends to zero as t ---+ 00.
(13) Let 1 < , < 3. Then, (3.13.151) is satisfied for (3.13.152) since
QT = 1, and
ft
19l(t)
ql(s)ds
r s-'Yds
lAt
_1_ (Al-'Y -1) tl-'Y

,-I
---+ 0 as t ---+ 00.
Therefore, by Theorem 3.13.28 there exists a decaying nonoscillatory solution of equation (3.13.152).
Oscillation and nonoscillation of haH~linear differential equations
267
Example 3.13.14. Consider again equation (3.13.129). The condition

(3.13.151) applied to (3.13.129) reduces to
k(J < _0'_;_1
(~) ",/(",+1)
or (J <
-,-(O'_+_l:...~l_/~(a~+~l_)
(a: 1)
",/(",+1) k- 1/(a+1).
(3.13.153)
It would be of interest to compare (3.13.153) with (3.13.133), which can be
rewritten as
This condition guarantees the nonexistence of bounded llonoscilla.tory solutions of (3.13.129).

1. The results of Sections 3.1 and 3.2 are based on the work of Li and
Yeh [40].
2. The results of Section 3.3 are extracted from Agarwal et. a1. [8] and
generalize a well-known theorem of Levin [38].
3. Theorem 3.4.1 is due to Hartman [25], also see Swanson [58]. All
other results in Section 3.4 are borrowed from Li and Yeh [45]. We note
that Theorem 3.4.5 is an extension of theorem 1 of Singh [57J.
4. Theorem 3.5.1 is taken from Duzurnak and Millgarilli [14], whereas
Lemmas 3.5.1 and 3.5.2 and Theorem 3.5.2 are adapted from Wong and
Yell [62J.
5. Section 3.6 contains the work of Chantladze et. a1. [12].
6. Theorems 3.7.1- 3.7.6 are due to Li [39]. However, Kusano and

Yoshida [36] have proved Theorem 3.7.5 earlier for the particular case
q1 (t) = q( t) by employing the Schauder-Tychonov fixed point theorem.
Theorem 3.7.6 is a generalization of theorem 3.5 in Kusano and Natito [34].
Theorems 3.7.7- 3.7.22 are modelled after Li and Yeh [41,42,46]. Corollaries 3.7.9 and 3.7.10 are theorems 3.3 and 3.4 of Kusano and Naito [34].
Theorems 3.7.23 and 3.7.24 are taken from Jingfa [26].
7. Theorems 3.8.1 ~ 3.8.4 are borrowed from Elbert [15,16]' Del Pino
et. a1. [56], Kusano et. a1. [33] and Kusano and Yoshida [36]. Theorems
3.8.5 and 3.8.6, respectively are extensions of Leighton's result [37] and
Wintener's criterion [60J.
8. The results in Subsection 3.9.1 are extensions of some of those established by Grace [21], Manojlovic [53J and Philos [55J. The results in
268
Chapter 3
Subsection 3.9.2 generalize the work of Kong [29]. The results in Subsection 3.9.3 are taken from Li and Yeh [42,43]. These results extend and
improve the work of Willett [59].
9. Section 3.10 contains the work of Li and Yeh [41-44,48]. Corollary
3.10.1 is a result due to Wintner [61] (see also theorem 2.17 in Swanson
[58]), whereas Theorems 3.10.3 and 3.10.4 extend and improve some results
ofYan [68].
10. Theorems 3.11.1 - 3.11.3 extend oscillation criteria of Kartsatos

[27], whereas Theorem 3.11.4 generalizes theorem 1 of Wong [63J.
11. The results in Section 3.12 are taken from Lian et. a.l. [49J and
extend the work of Harris and Kong [24J.
12. Lemma 3.13.1 is an extension of a result of Erbe [20], whereas
Lemma 3.13.3 generalizes a result of Kartsatos [28J. Lemma 3.13.4 and
its duality Lemma 3.13.5 are due to Koplatadze and Chanturia [30]' and
Lemma 3.13.6 is extracted from Philos [54J. Most of the results in Subsection 3.13.2 are the extensions of those established by Grace and Lalli [22J.
Lemma 3.13.7 is due to Mahfoud [50J. The results of Subsection 3.13.3 are
new. Subsection 3.13.4 is based on the work of Elbert and Kusano [19J.
Theorems 3.13.14 and 1.13.15 are due to Jingfa [26J. Lemma 3.13.9 and
Theorems 3.13.16 - 3.13.18 are taken from Kusano et. a1. [32J. Lemma
3.13.10 is due to Mahfoud [51J and Theorem 3.13.19 extends a result of
Mahfoud [51J. For a(t) == 1 the results of Subsections 3.13.6 and 3.13.7
reduce to those proved in Kusano and Lalli [31J.
13. For several other related works to this chapter see Agarwal et. a1.
[1-7], Bihari [10,11]' Dosly [13], Elbert [17,18]' Kusano et. a1. [35], Li and
Yeh [47], Mahfoud and Rankin [52], and Wong and Agarwal [64-67J.
3.15. References
1. R.P. Agarwal and S.R. Grace, Oscillations of forced functional dif-
ferential equations generated by advanced arguments, Aequationes lVlathematicae, to appear.

2. R.P. Agarwal and S.R. Grace, Oscillation criteria for second order halflinear differential equations with deviating arguments, Dyn. Cont. Disc.
Impul. Sys., to appear.
3. R.P. Agarwal and S.R. Grace, Oscillation of certain second order differential equations, Proc. Sixth Int. Conf. on Nonlinear Functional Analysis
and Applications, Korea 2000, Nova Science Publishers Inc., New York, to
appear.
4. R.P. Agarwal and S.R. Grace, Interval criteria for oscillation of second order half-linear ordinary differential equations, Functional Differential
Equations, to appear.
Oscilla,tion and nonoscillation of half-linear diHerential equations
269
6. R.P. Agarwal, S.R. Grace and D. O'Regan, Oscillation criteria for
certain nth order differential equations with deviating arguments, J. Math.
Anal. Appl., to appear.
7. R.P. Agarwal, S.R. Grace and D. O'Regan, On the oscillation of
second order functional differential equations, Advances Mathl. Sci. Appl.,
to appear.
8. R.P. Agarwal, W.-C. Lian and C.C. Yeh, Levin's comparison theorems for nonlinear second order differential equations, Applied Math. Letters 9(1996), 29-35.
9. A. Besicovitch, Almost Periodic Functions, Dover, New York, 1954.
10. I. Bihari, An oscillation theorem concerning the half-linear differential
equations of second order, Magyar Tud. Akad, lVlat. Kutato Int. Kozl.
8(1963), 275-280.
11. I. Bihari, Oscillation and monotonicity theorems concerning nonlinear differential equations of the second order, Acta Math. Sci. Hungar. 9(1968),
83-104.
12. T.A. Chanturia, N. Kandelaki and A. Lomtatidze, On zeros of solutions of a second order singular half-linear equation, Mem. Diff. Eqns.
Math. Phy. 17(1999), 127-154.
13. O. Dosly, Oscillation criteria for half-linear second order differential equations, Hiroshima Math. J. 28(1998), 507-521.
14. A. Dzurnak and A.B. Mingarelli, Sturm-Liouville equations with Besicovitch almost periodicity, Proc. Amer. Math. Soc. 106(1989), 647-653.
15. A. Elbert, A half-linear second order differential equation, in Qualitative Theory of Differential equations, Szeged-Societas Janos Bolyai, Colloq.
Math. Soc. Janos Bolyai 30, 1979, 153-180.
16. A. Elbert, Oscillation and nonoscillation theorems for some nonlinear ordinary differential equations, Lecture Notes in Math., 964, Springer-Verlag,
New York, 1982, 187-212.
17. A. Elbert, Asymptotic behavior of autonomous half-linear differential
systems on the plane, Studia Sci. Math. Hungar. 19(1984), 447-464.
18. A. Elbert, On the half-linear second order differential equations, Acta.
Math, Hungar. 49(1987), 487-508.
19. A. Elbert and T. Kusano, Oscillation and nonoscillation theorems for a
class of second order quasilinear differential equations, Acta. Math. Hungar, 56(1990), 325-336.
20. L.H. Erbe, Oscillation criteria for second order nonlinear delay equations,
Ganad. Math. Bull. 16(1973), 49-56.
22. S.R. Grace and B.S. Lalli, Oscillatory behavior for nonlinear second
order functional differential equations with deviating arguments, Bull. Jnst.
Math. Aad. Sinica 14(1986), 187-196.
23. G.H. Hardy, J .E. Littlewood and G. Polya, Inequalities, 2nd ed.
Cambridge Univ. Press, 1988.
270
Chapter 3
24. B.J. Harris and Q. Kong, On the oscillation of differential equations

with an oscillatory coefficient, Trans. Amer. Math. Soc. 347(1995), 18311839.
1964.
26. W. Jingfa, On second order quasilinear oscillations, Funkcial. EkYac.
41(1998), 25--54.
27. A.G. Kartsatos, On oscillations of nonlinear equations of second order,
J. Math. Anal. Appl. 24(1968), 357-361.
28. A.G. Kartsatos, On nth order differential inequalities, J. Math. Anal.
Appl. 52(1975), 1-9.
29. Q. Kong, Interval criteria for oscillation of second order linear ordinary
differential equations, J. Math. Anal. Appl. 239(1999), 258-270.
30. R.G. Koplatadze and T.A. Chanturia, On oscillatory and monotone
solutions of first order differential equations with deviating arguments, Differencial'nye Uraynenija 18(1982), 1463-1465.
31. T. Kusano and B.S. Lalli, On oscillation of half-linear functional differential equations with deviating arguments, Hiroshima Math. J. 24(1994),
549-563.
32. T. Kusano, A. Ogata and H. Usami, Oscillation theory for a class of
second order quasilinear ordinary differential equations with application to
partial differential equations, Japan J. Math. 19(1993), 131-147.
33. T. Kusano, A. Ogata and H. Usami, On the oscillation of solutions of
second order quasilinear ordinary differential equations, Hiroshima Math.
J. 23(1993), 645-667.
34. T. Kusano and Y. Naito, Oscillation and nonoscillation criteria for second order quasilinear differential equations, Acta Math. Hungar. 76(1997),
81-99.
35. T. Kusano, Y. Naito and A. Ogata, Strong condition and nonoscillation of quasilinear differential equations of second order, Dilf. Eqns. Dyn.
Sys. 2(1994),1-10.
36. T. Kusano and N. Yoshida, Nonoscillation theorems for a class of
quasilinear differential equations of second order, J. Math. Anal. Appl.
189(1995), 115-127.
37. W. Leighton, The detection of the oscillation of solutions of a second
order linear differential equation, Duke J. Math. 17(1950), 57-61.
38. A.Y. Levin, A comparison principle for second order differential equations, SOY. Math. DoH 1(1960), 1313-1316.
39. H.J. Li, Nonoscillation characterization of second order linear differential
equations, Math. Nachr. 219(2000), 147-161.
40. H.J. Li and C.C. Yeh, Sturmian comparison theorem for half linear
second order differential equations, Proc. Roy. Soc. Edinburgh Sect. A
125(1995), 1193-1204.
41. H.J. Li and C.C. Yeh, Nonoscillation criteria for second order half linear
differential equations, Appl. Math. Letters 8(1995), 63-70.
271
42. H.J. Li and C.C. Yeh, Nonoscillation theorems for second order quasilinear differential equations, PubJ. l\1ath. (Debrecen) 47(1995), 271-279.
43. H.J. Li and C.C. Yeh, Oscillation of half linear second order differential
equations, Hiroshima Math. J. 25(1995), 585-594.
44. H.J. Li and C.C. Yeh, Oscillation criteria for nonlinear differential equations, Houston J. Math. 21(1995), 801811.
45. H.J. Li and C.C. Yeh, An oscillation criterion of almost periodic SturmLiouville equations, Rocky Mount. J. Math. 25(1995),1417-1429.
46. H.J. Li and C.C. Yeh, On the nonoscillatory behavior of solutions of a
second order linear differential equation, Math. Nachr. 182(1996), 295315.
47. H.J. Li and C.C. Yeh, Oscillation of nonlinear functional differential
equations of the second order, AppJ. Math. Letters 11(1)(1998), 71-77.
48. H.J. Li and C.C. Yeh, Oscillation and nonoscillation criteria for second
order linear differential equations, Math. Nachr. 194(1998),171-184.
49. W.C. Lian, C.C. Yeh and H.J. Li, The distance between zeros of an
oscillatory solution to a half linear differential equation, Comput. Math.
Appl. 29(1995), 39-43.
50. W.E. Mahfoud, Oscillation and asymptotic behavior of solutions of nth
order nonlinear delay differential equations, J. Differential Equations 24
(1977), 75-98.
51. W.E. Mahfoud, Comparison theorems for delay differential equations,
Pacific J. Math. 83(1979), 187-197.
52. W.E. Mahfoud and S.M. Rankin, Some properties of solutions of
(r(t)lJ!(x)x')' + a(t)f(x) = 0, SIAM J. Math. Anal. 10(1979),49-54.
53. J.V. Manojlovic, Oscillation criteria for a second order half-linear differential equation, Math. Comput. Modelling 30(5-6)(1999),109--119.
54. Ch.G. Philos, On the existence of nonoscillatory solutions tending to
zero at 00 for differential equations with positive delays, Arch. Ivlath.
36(1981),168-178.
55. Ch.G. Philos, Oscillation theorems for linear differential equations of second order, Arch. Math. 53(1989), 482-492.
56. M. Del Pino and R. Manasevich, Oscillation and nonoscillation for
(lu'I P - 2 u')' + a(t)luI P - 2 u = 0, p> 1, Houston J. Math. 14(1988), 173177.
57. B. Singh, Comparative study of asymptotic nonoscillation and quickly
oscillation of second order linear differential equations, J. Ivlathl. Phyl.
Sci. 4(1974), 363-376.
59. D. Willett, On the oscillatory behavior of the solutions of second order
linear differential equations, Ann. Polon. Math. 21(1969), 175-194.
60. A. Wintner, A criterion for oscillatory stability, Quart. Appl. Math.
7(1949), 115-117.
61. A. Wintner, On the nonexistence of conjugate points, Amer. J. Math.
73(1951), 368-380.
272
Chapter 3
62. F.R. Wong and C.C. Yeh, An oscillation criterion for Sturm-Liouville
equations with Besicovitch almost periodic coefficients, Hiroshima Math.
J. 21(1991), 521-528.
63. J .S.W. Wong, Oscillation criteria for a forced second order linear differential equation, J. Math. Ana1. App1. 231(1999), 235-240.
64. P.J.Y. Wong and R.P. Agarwal, Oscillation theorems and existence
criteria of asymptotically monotone solutions for second order differential
equations, Dynam. Systems App1. 4(1995), 1177-496.
65. P.J.Y. Wong and R.P. Agarwal, On the oscillation and asymptotically
monotone solutions of second order quasilinear differential equations, Appl.
Math. Comput. 79(1996), 207-237.
66. P.J.Y. Wong and R.P. Agarwal, Oscillatory behavior of solutions of
certain second order nonlinear differential equations, J. Ivlath. Ana1. Appl.
198(1996), 813-829.
67. P.J.Y. Wong and R.P. Agarwal, Oscillation criteria for half-linear differential equations, Advances Math. Sci. Appl. 9(1999), 649-663.
68. J. Yan, Oscillation property for second order differential equations with
an 'integral small' coefficient, Acta Math. Sinica 30(1987), 206--215.
Chapter 4
Oscillation Theory for
Superlinear Differential
Equations
4.0. Introduction
This chapter presents oscillation and nonoscillation theory for solutions
of second order nonlinear differential equations of superlinear type. Section
4.1 deals with the oscillation of superlinear equations with sign changing
coefficients. Here, first we shall discuss some results which involve integrals
and weighted integrals of the alternating coefficients, and then provide several criteria which use average behaviors of these integrals. More general
averages such as 'weighted averages' and 'iterated averages' are also employed. In Section 4.2, we impose some additional conditions on the superlinear terms which allow us to proceed further and extend and improve
some of the results established in Section 4.1. In fact, an asymptotic study
has been made and interesting oscillation criteria have been proved. In
Section 4.3, first we shall provide sufficient conditions which guarantee the
existence of nonoscillatory solutions, and then present necessary and sufficient conditions for the oscillation of superlinear equations. Oscillation
results via comparison of nonlinear equations of the same form as well as
with linear ones of the same order are also established. In Section 4.4,
we shall extend some of the results of the previous sections and establish
several new oscillation criteria for more general superlinear equations. Necessary and sufficient conditions for such equations to be oscillatory are also
given. Section 4.5 deals with the oscillation of forced-super linear differential equations with alternating coefficients. Finally, Section 4.6 presents the
oscillation and nonoscillation criteria for second order superlinear equations
with nonlinear damping terms.
4.1. Superlinear Oscillation Criteria

In this section we shall present oscillation criteria for superlinear ordi-
274
Chapter 4
nary differential equations of the form
J;"(t)
+ q(t)f(x(t))
0,
+ q(t)f(x(t))
(a(t)x'(t))'
(4.1.1 )
(4.1.2)
and the more general damped equation
(a(t)x'(t))'
+ p(t)x'(t) + q(t)f(x(t))
0,
(4.1.3)
where
a(t) E C([to, (0), m+),

p(t), q(t) E C([to, (0), JR),
(i)
(ii)
(iii) f E C(JR, JR),
and satisfies
f is continuously differentiable except possibly at 0,
xf(x) > 0 and

also
f'(x) 2
for
x =I 0
( 4.1.4)
<
(4.1.5)
f (x) is strongly superlinear in the sense that
-= du
f(u)
The special case when f(x) = Ixl 7 sgn x, x E JR

interest. In fact, the differential equations
x"(t)
+ q(t)lx(t)17
(a(x)x'(t))'
and
(a(t)x'(t))'
sgn x(t)
+ q(t)lx(t)17
1) is of particular
= 0,
sgn x(t)
+ p(t)x'(t) + q(t)lx(t)17
b >
00.
(4.1.6)
= 0
sgn x(t)
(4.1.7)
(4.1.8)
arc prototype of (4.1.1), (4.1.2) and (4.1.3) respectively, and will be discussed extensively.
We begin with the following oscillation criteria for equation (4.1.2).
Theorem 4.1.1.
Suppose conditions (4.1.4) and (4.1.5) hold,
1=
to
L
OO
lim inf
t--+CXl
-ds =
a( s)
00,
(4.1.9)
q(s)ds <
00,
(4.1.10)
J7fTt q( s )ds
for all large
(4.1.11)
Oscillation theory for superlinear differential equations
i 1
and
00
lim t -1()
q( u )duds
t--->oo to a s s
275
(4.1.12)
00.

Proof. Let x(t) be a nonoscillatory solution of (4.1.2), say, x(t) > 0
for t:::: to:::: O. Define wet) = a(t)x'(t)/f(x(t)), t:::: t 1 . Then, we have
w'(t)
T:::: tl
For
q(t) -
f'~~g)) w 2(t) ~
- q(t)
an integration of (4.1.13) yields
a(t)x'(t)
a(T)x'(T)
f(x(t)) ~ f(x(T)) -
t:::: t 1 .
for
t ( )d
iT q
(4.1.13)
s.
Now, we need to consider the following cases:

Case I.
If x'(t) :::: 0 for all
o ~
t:::: T,
then from (4.1.10), we find
a(T)x'(T)
f(x(T)) -
roo
iT
q(s)ds.
The same type of reasoning as above guarantees for t:::: T that we have
ftOO q(s)ds ~ a(t)x'(t)/ f(x(t)). Dividing both sides of this inequality by
aCt) and integrating from T to t, we get
it 1
1
00
a(s)
q(u)duds <
-
it
T
l (t)
x'(s)
ds =
f(x(s))
x(T)
de,
.
f(e,)
This in view of (4.1.12) contradicts condition (4.1.5).

Case II. If x'(t) changes sign, then there exists a sequence {Tn};:"=l'
limn--->oo Tn = 00 such that x'(Tn) < o. Choose N large enough so that
(4.1.11) holds. Then, we have
a(t)x'(t) < a(TN)x'(TN) f(x(t)) f(x(TN))
it
TN
and hence
a(t)x'(t)
h~~p f(x(t))
a(TN)x'(TN)
f(x(TN))
+ li~s~p
()d s,
qs
( tq(s)ds
)
-
iTN
< 0,
which contradicts the fact that x'(t) oscillates.

Case III. If x'(t) < 0 for t:::: t2 :::: tI, then condition (4.1.11) implies
that for To :::: to there exists T 1 :::: To such that f;, q(s)ds :::: 0 for all
t :::: T 1 . Choosing Tl :::: t2 as indicated and then integrating (4.1.2) from
Chapter 4
276
TI to t, we find
a{t)x'(t)
= a(Tdx'(Td - t q(s)f(x(sds
iT!
= a(TI)x'(TI)- f(x(t)) (tq(s)ds+ tf'(x(sx'(s)
iT,
iT!
iT,
q(u)duds
S a(T1)x'(TI).
Thus, x(t) S X(TI) + a(TI)x'(TI) J;! dsja(s) ---+ -00 as t ---+ 00, which
contradicts the fact that x(t) > 0 for t ~ i}. A similar proof holds when
x(t) <0 for t~tl'
Remark 4.1.1. The condition (4.1.11) on the coefficient q(t) is equivalent

to the following: either
lim l t q(s)ds =
t--+oo to
(4.1.11h
00,
or limt--+oo Jt: q( s )ds exists and is finite, and

Q(t)
00
q(s)ds
for all large t.
(4.1.11h
For this, let (4.1.11) hold. Define Q*(t) = Jt: q(s)ds. We first claim that
liminft--+oo Q*(t) i= -00. Suppose this is not true. Then there exists a
sequence {tn}~=l' tn ---+ 00 as n ---+ 00 such that Q*(tn ) S -no For
large T, wehave J;nq(s)ds=Q*(tn)-Q*(T)S-n-Q*(T)---+-oo as
t ---+ 00, which contradicts (4.1.11). If liminft--+oo Q*(t) = 00, we have
(4.1.11h. Thus, we assume that liminfQ*(t) = a i= 00. Suppose that
t--+oo
limsuPt--+oo Q*(t) = f3 > a. Then there exist two sequences {tn}~=l and
{Tn}~=l' tn ---+ 00 and Tn ---+ 00 as n ---+ 00 such that Q*(tn ) ---+ a
and Q*(Tn) ---+ f3 as n ---+ 00. Then, J;~ q(s)ds = Q*(t m ) - Q*(Tn) <
(f3-a)j2 (with appropriate modification if f3 = 00) for m, n sufficiently
large. This contradicts (4.1.11). Thus, limt--+oo Q*(t) exists, and Q(T) =
limt--+oo J; q(s)ds = liminft--+oo J; q(s)ds ~ 0 as assumed in (4.1.11). The
sufficiency part is obvious.
Theorem 4.1.2. Let a(t) Sal, t ~ to where al is a positive constant,
and condition (4.1.5) holds. If for every large T there exists a TI ~ T
such that
lim inf
t--+oo
and
t r q(u )duds
iT! iT
00
to
sq(s)ds
> 00,
00
(4.1.14)
(4.1.15)
277

~ to. Define w(t) as in Theorem 4.1.1 and obtain (4.1.13).
Next, we need to consider the following three cases:
Proof.
x(t) > 0 for t
Case 1. If x'(t) oscillates, choose T ~ to so that x'(T) = 0 and (4.1.14)

holds, then integrating inequality (4.1.13), we get a(t)x'(t)j f(x(t)) :S;
- f;q(s)ds. It follows from (4.1.14) that there exist Tl ~ T and A> 0
such that
t a(s)x'(s)
iTt f(x(s)) ds
:S;
t r
iTt iT q(u)duds
< A for all
~ T1
Multiplying both sides of (4.1.13) by t and integrating from Tl to t,

we obtain
ta(t)x'(t)
T1a(Tdx'(Tt)
f(x(t)) :S;
f(x(T1))
+ -
{t
()d
iTt sq S
s,
which by (4.1.15) contradicts the assumption that x'(t) oscillates.

Case II. If x'(t) > 0 for t ~ h ~ to, then multiplying (4.1.13) by t
and integrating from tl to t, we obtain
ta(t)x'(t)
f(x(t))
t1a(t1)x'(tt) + it a(s)x'(S)ds - it sq ()d

s s
< -:-':--':--:-':---'f(x(td)
tla(tl)x'(td
f( X (tl ))
t, f(x(s))
+ al
t,
(t) du
it
f( ) sq(s)ds,
x(t,)
t,
which in view of (4.1.5) again leads to a contradiction.

Case III. If x'(t) < 0 for t ~ t2 ~ tl condition (4.1.15) implies that there
exists a t3 ~ t2 such that ft:3 sq(s)ds = 0 and ftt3 sq(s)ds ~ 0 for
t ~ t3. Multiplying (4.1.2) by t and integrating by parts from t3 to t,
we obtain
ta(t)x'(t)
< t3a(t3)X'(t3) - i t sq(s)f(x(s))ds

t3
< t3a(t3)X'(t3) - f(x(t)) it sq(s)ds

t3
(t f'(x(s))x/(s) i
t3
uq(u)duds :S; ha(t3)x'(t3),
t3
and hence x'(t):S; t3a(t3)X/(t3)j(alt) for t
x(t) :S; X(t3)
+ t3 a(t3)X'(t3) In!...
al
t3
t3. Thus, we have
--+ -
00
as
t --+
00
Chapter 4
278
which contradicts the fact that x(t) > 0 for t
to.
Remark 4.1.2. We note that the condition a(t) ~ a} for t ~ to in

Theorem 4.1.2 can be replaced by a'(t) ~ 0 for t ~ to. In this case we
need to apply the Bonnet theorem to obtain the desired conclusion.
Theorem 4.1.3. Suppose there exists a function p(t) E C 2 ([to, (0), JR+)
such that
1
1
00
and
00
p(s)q(s)ds =
00
(4.1.16)
00.
(4.1.17)
--:--:--:---:- ds
a(s)p(s)
Then the following hold:

(i}) If fJ(t) = a(t)p'(t) - p(t)p(t) ~ 0 and fJ'(t) ~ 0 for t ~ to, then all
bounded solutions of (4.1.3) are oscillatory. If in addition condition (4.1.5)
holds, then (4.1.3) is oscillatory.
(i2) If
Joo
du
-- <
+0 f(u)
00
and
du
Lo f(u)
<
(4.1.18)
00
hold, and
IfJ'(s)lds < 00, then all bounded solutions of (4.1.3) are
oscillatory. If in addition condition (4.1.5) holds, then (4.1.3) is oscillatory.
~ to. Define w(t) = p(t)a(t)x'(t)/ f(x(t)), t ~ t} for some
to. Then for t ~ t}, we have
Proof.
x(t) > 0 for t
t} ~
w'(t)
-p(t)q(t)
x'(t)
+ f3(t) f(x(t))
1
2('( ( ))
- a(t)p(t) w t)f x t
x'(t)
< -p(t)q(t) + f3(t) f(x(t))'

Integrating this inequality from t} to t, we obtain
w(t) ~ w(t}) -
x'(s)
it}r p(s)q(s)ds + itr1f3(s) f(x(s))
ds.
Now, we shall show that each of the conditions (h) and (i2) ensures
that
f3(s)x'(s)/f(x(s))ds is bounded above. Consider the following
two cases:
Jtt}
I}. Let (i}) hold. By the Bonnet theorem, for a fixed t ~ tl and ~ E [t}, t),
we have
x'(s)
it 1f3(s) f(x(s)) ds
($. x'(s)
= fJ(tI) i h f(x(s)) ds
r($.)
f3(tI)
iX(tl)
du
f(u)
279
and hence, since (3(t)?: 0, and
we find
[
where KI
t,
f(u) <
x(ttl
if x(~) < x(t l )
du
()
00
x(t,)
du
-f()
u
x'(s)
(3(s) f(x(s ds < KI
(4.1.19)
= (3(tl ) Ix'7't>l du/ f(u).
p(s)q(s)ds :s: - w(t)
t,
t?: tl,
for all
From (4.1.19) it follows that

[
x(~)?: x(td,
if
+ w(t l ) + KI
[t
t,
( ) ( ) wZ(s)f'(x(sds.
asp s
(4.1.20)
12 . Let (iz) hold. Then, we have

t
t,
x'(s)
(3(s) f(x(s ds
ret)
= (3(t) io
-
t,
du
ret,) du
f(u) - (3(td io
f(u)
(r(S) f(u)
du )
(3'(s) io
ds.
(4.1.21)
I;(tJ
In this case, it follows from (iz) that O:s:

du/ f(u) :s: M and
1(3(t) I :s: N, t ?: tl where M and N are real constants. Also, the
integral Itt, (3'(s)
du/ f(u) ds converges absolutely as t -t 00 and
(I;(S)
so, (4.1.21) implies

[
t,
x'(s)
(3(s) f(x(s ds :s: MI
where MI = (3(td
t,
+ MIN + M
I;(t,) du/f(u).
p(s)q(s)ds :; - w(t)
[00 1(3'(s)lds
t,
Thus, we obtain
+ W(tl) + Kz -
it
t,
( ) ( ) wZ(s)f'(x(sds.
asp s
(4.1.22)
Set K = max{KI, K z }. Then, (4.1.20) and (4.1.22) can be written as
t
t,
p(s)q(s)ds :s: - w(t)
+ w(td + K
-it
t,
()1 ( ) wZ(s)f'(x(sds.
asp s
(4.1.23)
280
Chapter 4
It follows from condition (4.1.16) that limHoo p(t)a(t)x'(t)/ f(x(t)) =

-00,
i.e.
x'(t) < 0 for t::::: t2 ::::: tl. Let t3::::: t2 be such that
W(tI) + K p(s)q(s)ds <::: -1 for t::::: t 3, then (4.1.23) implies
Ittl
-w(t) ::::: 1 +
1 () ()
1
tl
asp s
w 2(s)f'(:r(s))ds,
(4.1.24)
or
1l.,2(t)f'(:r(t))/a(t)p(t)
>
x'(t) ,
f(x(t)/
(x(t))
1 -+ 1/1 (w 2(S)f'(x(s))/a(s)p(s)) ds
and consequently for all t::::: t3,
In{~ (1+
where )..
1+
It
(w 2(s)f'(x(s))/a(s)p(s))dS)} :::::
In ~(~(;:/;,
= 1 + Ittl 3 (w 2(s)f'(x(s))/a(s)p(s)) ds. Hence, it follows that
1( 2 '
t
tl
)
f(X(t3))
W (s)f (x(s))/a(s)p(s) ds ::::: ).. f(x(t))
Thus, (4.1.24) yields x'(t) <::: -)..f(X(t3))/a(t)p(t) for t::::: t3. Therefore,
we have
t
1
x(t) <::: X(t3) - [).. f(X(t3))]
1
t3
( ) ( ) ds
asp s
which in view of (4.1.17) leads to a contradiction that limHoo x(t) =
-00.
Finally, we remark that in the above proof the oscillation of all bounded
solutions of (4.1.3) is trivially included.
The following immediate corollaries of Theorem 4.1.3 are well known

basic results in the literature.
Corollary 4.1.1. If 'Y > 1, and
IX! sq(s)ds
00,
(4.1.25)

Corollary 4.1.2. If conditions (4.1.5) and (4.1.25) hold, then equation
Corollary 4.1.3. If condition (4.1.5) holds,
ta )' >0
( -a(t)
-,

where 0 <
281
S 1, then equation (4.l.2) is oscillatory.
= tC</a(t), t 2" to and
Proof. In Theorem 4.1.3, let p(t) == 0 and p(t)

O<oSl.
(~X'(t))' + t~x'(t)+ [-C1/2sint+~C3/2(2+Cost)]
(4.l.26)
= 0, t 2" to = n/2
l. Let p(t) = t. Then, j3(t) = a(t)p'(t) -- p(t)p(t) = 2/t and
x (lx(t)I"Y) sgn :r(t)
where 'Y>
for every t 2" to, we have
t
to
sq(s)ds
j
j
~s-1/2(2+COss)J
t
n/2
[-sl/2sins+
t
7r/2
d[sl/2(2+coss)]
> t 1/ 2
2( n /2) 1/2
-7 00
ds
t 1/ 2(2+cost)-2(n/2)1/2
as t
-7 00.
Now, Theorem 4.l.3(h) ensures that (4.l.26) is oscillatory.

xll(t)+~:r'(t)+ [_t1/2 sint + ~C3/2(2 + cost)]
f(x(t)) = 0, t 2" to = n/2
where f E C(lR, lR), f(x) = Fx(I + x) for x>

Here, we have
X(t)
du
y'U(I
+ u)
= 2 tan -1 Vx and so
(4.1.27)
and - f( -x) = f(x).
OO du
f (u)
<
00.
Let p(t) = t. Then, j3(t) = 1- (I/t). Hence, Theorem 4.1.3(i 2) guarantees

that (4.l.27) is oscillatory. We also note that Theorem 4.1.3(h) is not
applicable to (4.l.27).
An interesting result for (4.l.2) which follows from Theorem 4.l.3 by
taking p( t) = 6 (t) is the following:
fX)
It:
Corollary 4.1.4. If condition (4.l.9) holds, and

6(s)q(s)ds =
where 6(t) =
ds/a(s), then equation (4.l.2) is oscillatory.
00,
Now, we shall prove the following result.

Theorem 4.1.4. Suppose condition (4.l.5) holds and there exists a real
constant k such that
j'(x) 2" k >
for
=1=
0.
(4.1.28)
282
Cha.pter 4
Let p(t) E C 1 ([to,oo),]R+) be such that condition (4.1.17) hold,
(3(t)
a(t)p'(t) - p(t)p(t) ~ 0 and (3'(t) :::; 0 for
t ~ to (4.1.29)
and
1 it a(s)p(s)ds <
lim sup "2
If
t ....= t
to
li~~f
p(s)q(s)ds > -
1:
00.
(4.1.30)
00
(4.1.31)
and
lim sup -l i t i s p(u)q(u)duds
t ....
to
(4.1.32)
00,
to

x(t) > 0 for t ~ to. Define w(t) = p(t)a(t)x'(t)/f(x(t)), t ~ to. Then
for every t ~ to, we have
w'(t) =
Hence, for all t
- p(t)q(t)
~
+ (3(t) f~~~ti))
- a(t)lp(t) w 2 (t)f'(x(t)).
to, it follows that
i to p(s)q(s)ds = - w(t)
+ w(to) +
By the Bonnet theorem, for any t
it (3(s)x'(s)
it w 2 (s)f'(x(s))
to f(x(s)) ds to
a(s)p(s) ds.
to there exists a
t
x'(s)
i x'(s)
i to (3(s) f(x(s)) ds = (3(to) to f(x(s)) ds
~ E
(3(to)
< (3(to)
Thus, for every t
[to, t] such that
du
x(to} f(u)
x (}
00
x(to}
du ).
-f(
U
to, we find
t ( ) ( )d
() M - it w 2 (s)f'(X(S))d s
i to psqs s:::; -wt+
to
a(s)p(s)
,
(4.1.33)
where M = w(to)+(3(to) f*o} duj f(u). Therefore, from condition (4.1.28),

we get
it
~
p(s)q(s)ds :::; -w(t)+M -k
it
w 2 (s)
( ) ()ds for all t ~ to. (4.1.34)
~asps
283
Next, we need to consider the following three cases for the behavior of x'(t).
Case 1. x'(t) is oscillatory. Then there exists a sequence {tm}~=l in
[to, 00) with limm-too tm = 00 such that x'(tm) = 0, m = 1,2,.
Thus, (4.1.34) gives
t",
to
11
W2(S)
M
()()ds:::;-k--k
asp s
,
tm
to
p(s)q(s)ds,
m=1,2,
and hence, in view of (4.1.31),
00
to
w2(s)
-:-:-'---'--,-ds <
a( s )p( s)
(4.1.35)
00.
It:
w 2(s)/(a(s)p(s))ds :::; N for every t

Therefore, for some N > 0,
By the Schwartz inequality, for t ~ to, we have
I-l
W(S)dSI'
~ 11: Ja(.,)p(.,) [Ja7:;~(s) 1d{
: :; [1:
a(s)p(s)dsJ
1: a~~~~~)dS
From condition (4.1.30) there exists a
Ct 2 for t ~ to,
to
-It
p(u)q(u)duds :::;
to
to
p(s)q(s)ds :::; -w(t)
It: w(s)ds
+M
a(s)p(s)ds.
:::; VNCt.
for t ~ to, and
w(s)ds+M(t-to) :::; (VNC+M)t-Mto,
to
litis
i.e.,
Itto
to.
It: a(s)p(s)ds :::;
0 such that
and hence for all t ~ to,
Furthermore, (4.1.34) gives

hence for all t ~ to,
I 15
c>
1:
:::; N
p(u)q(u)duds :::; VNC + M _Mt

_0.
t
to
This contradicts condition (4.1.32).

Case 2. x'(t) > 0 on [T,oo) for some T ~ to. In this case, from (4.1.34)
it follows that for t ~ T,
p(s)q(s)ds :::; M, and consequently
It:
!
t
rt 15 p(u)q(u)duds
iT
to
:::; M
(1 -~)t
for
which again contradicts condition (4.1.32).

Case 3. x'(t) < 0 on [T,oo) for some T
I -
00
to
to. If
w2(S)!'(X(S))d
s <
a(s)p(s)
00
~ T,
Chapter 4
284
then condition (4.1.28) ensures that (4.1.35) holds. But, then as Case 1 we
get a contradiction. Thus, we assume that I = 00. Now, exactly as in
the last part of the proof of Theorem 4.1.3, we arrive at the contradiction
limHoo x(t) = -00.
In the following result we shall employ the averaging technique and the
weighted integral functions to discuss the oscillatory behavior of equation
(4.1.3).
Theorem 4.1.5. Let conditions (4.1.5) and (4.1.28) hold, and assume
that there exist p(t) E C1([to, 00), m+) and cf; E (t, to), where (t, to)
denotes the class of all positive and locally integrable functions on [to, 00)
such that (4.1.29) is satisfied,
(It:
cf;(s)
cf;(u)du) >.
rS ()
)
2(
)
ds
lto au p(u cf; u) du
00
/
is
and
lim
t---+oo fto
it cf;(s)
cf;( s )ds
to
for some constant A, 0 < A < 1
00
(4.1.36)
p(u)q(u)duds
00.
(4.1.37)
to

x(t) > 0 for t ~ to. Define w(t) = p(t)a(t)x'(t)/f(x(t for t ~ to
Proof.
and proceed as in Theorem 4.1.4 to obtain (4.1.34). Multiplying (4.1.34)

by cf;(t) and integrating from to to t, we get
to
is
cf;(S)W(S)ds+klt cf;(s)
to
to
~~(~)
t
::; Mit cf;(S)dS-l cf;(s)
to
to
Using condition (4.1.37), there exists a tl
M -
fto
Then for all t
o ::;
cf;( s )ds
F(t)
it is
cf;(s)
to
)
is
a u Pu
duds
(4.1.38)
p(u)q(u)duds.
to
to such that
p(u)q(u)duds < 0 for
tl, it follows that
= k
it is
to
cf;(s)
tl
to
to
w 2 (u)
( ) ( ) duds < a u Pu
it
to
cf;(s)w(s)ds
and
(4.1.39)
285
By the Schwartz inequality, we have
[ltDt
[4>(s)Ja(s)p(s)] (
w (s)
)
(ltDt a(s)p(s)l(s)ds) (rltDt a(s)p(s)
ds
4>tt)
P(t) :2: k
(4.1.40)
<
We also have
w(s) ) dS]2
Ja(s)p(s)
(1:
a(s)p(s)4>2(S)dS) P'(t)
i (itl
t
t,
4>(s)
for
t;::: tl
it
2
W
( )(U))
( ) du ds = c
4>(s)ds,
a u pUt,
(4.1.41)
where c = k It:'w2(s)/a(s)p(s)ds. Now, from (4.1.39), (4.1.40) and (4.1.41),

we obtain
cA 4>(t)
(Itt 4>( s )ds ) A

'
fto a(s)p(s)4>2(s)ds
::;
p A - 2 (t)P'(t)
for all
t;::: t l ,
(4.1.42)
where 0 < A < 1. Integrating (4.1.42) from tl to t, we get

cA
t,
4>(s)
(It: 4>(u)du) A l l
ds < - A
<
Ito a(u)p(u)4>2 (u)du
1 - A pl- (tI)
s
00,
Remark 4.1.3.
Condition (4.1.28) fails in the special case f(x) =
lxi' sgn x, x E JR b > 1). However, it is satisfied for the following
functions:
(I)
(II)
f(x) =
f(x) =
(III) f(x) =
(IV) f(x) =
mx + lxi' sgn x, x E 1R. for 'Y > 1 and any constant m> 0,
xln2(p + Ix!), x E JR for any constant p> 1,
xe AlxI , x E lR for any constant A> 0,
sinh x, x E JR.
In our next result we do not need condition (4.1.5).
Theorem 4.1.6. Let condition (4.1.28) hold, p(t) E Cl([to,oo),lR+),

4>(t) E iI>(t, to) be such that [3(t) = a(t)p'(t) - p(t)p(t) :2:
for t;::: to,
and condition (4.1.36) is satisfied. If
it it [
.
1
1 [32 (u) ]
lIm t
4>(s)
p(u)q(u) - k ( ) ( ) duds
t ..... oo fto 4>(s)ds to
to
4 au p u
00,
(4.1.43)
Chapter 4
286
Proof.
x(t) >
t ?: to,

t?: to. We define w(t) as in Theorem 4.1.3 and obtain for
for
(3(t)
1
2
,
-p(t)q(t) + a(t)p(t) w(t) - a(t)p(t) w (t)f (x(t
w'(t)
k
[
(3(t)] 2
1 (32(t)]
< - [p(t)q(t) - 4k a(t)p(t) - a(t)p(t) w(t) - 2k
Hence, we have
w(t)
+k
t
1
[
(3(s)] 2
to a(s)p(s) w(s) - 2k ds
-s: w(t o) -
t [
1 (32(s) ]
to p(s)q(s) - 4k a(s)p(s) ds.
The rest of the proof is similar to that of Theorem 4.1.5.
Corollary 4.1.5. In Theorem 4.1.6, condition (4.1.43) can be replaced by

(4.1.37) and
oo
(32(S)
a(s)p(s) ds <
(4.1.44)
00.
C~X/(t)'+S~~txl(t)+~(~-sint)f(x(t
t>to=l (4.1.45)
0,
where the function f is as in Remark 4.1.2 which satisfies condition

(4.1.28). Let p(t) = (t) = t. Then, we have
l:
t (32(s) d
to a(s)p(s) s
p(S)q(S)dS
{t (1- sins)2 d
= Jl
it(~-sins)ds
s3
it (s) 1to p(u)q(u)duds?:

fto(s)ds to
1
<
00,
Int+cost-cos1 > Int-2, t>l
and
1-2
t -1
jt
s[lns - 2Jds
-t 00
as t -t 00. Thus, all the conditions of Corollary 4.1.5 are satisfied with
A E (0,1), where A is the constant in (4.1.36), and hence (4.1.45) is
oscillatory.
Oscillation theory for superlincar differential equations
287
(t 2 x'(t))' - 2tx'(t)
20 + cost + -2-(2
0, t 2:: to> 7r/2,
2tsint)f(x(t))
(4.1.46)
where f(x) is as in Remark 4.l.2. We take pet) = 1/t 2 and (t) = t, t 2::
to> 7r/2. Then, (3(t) = a(t)p'(t) - p(t)p(t) = 0 for t > 7r/2. Further,
1
t
p(s)q(s)(1.s
to
([-~Sin8+ 2ys
\:;(2+COSS)]dS
./7r/2
./7r /2
Vi(2
and
t 1
1 t ( s)
Ito ( s )ds to
1
8
to
d [vIS(2
+ coss)]
+ cost)
- 2J(7r/2) >
p( v.)q( 11.)d11.ds 2::
t - 7r
/2)21
Vi -
7r /2
-+
2J(7r/2)
(vis - Vq)
ds
"2
2
00
as
t -+
00.
Thus, all the hypotheses of Corollary 4.l.5 are satisfied with A E [1/2,1) C
(0,1), and hence (4.l.46) is oscillatory.
Now, we shall employ some integral inequalities to study the oscillation of some of the above equatiolls. For this, we shall lleed the following
lemmas.
Lemma 4.1.1. Let k(t, s, y) be a real-valued function of t, s in [b, c)
and y in [bl, CI) such that for fixed t = to and s = So, keto, So, y) is a
nondecreasing function of y. Let G(t) be a given function on [b, c); let
v. and v be functions on [b, c) such that 11.( s) and v( s) are in [b l , CI)
for all s in [b,c); let k(t,s,v(s)) and k(t,s,u(s)) be locally integrable
in s for fixed t; and for all t E [b, c). Let
vet) = G(t)
it
k(t,s,v(s))ds
and
u(t) 2:: G(t)
it
k(t,s,11.(s))ds.
Then, v(t):S 11.(t) for all t E [b, c).

Lemma 4.1.2. Let x(t) be a positive (negative) solution of equation
(4. l.2) on [b l , c) for some positive number bi satisfying to :S bi < c :S 00.
If there exist b E [b l , c) and a positive constant C such that
-w(b l
1
t
b,
q(s)ds
1b -(
b,
1
)w 2(s)f'(x(s))ds 2:: C,
a s
(4.1.47)
Chapter 4
288
w(t) = a(t)x'(t)/ f(x(t)) for all t E [b, c),

-Cf(x(b)) (a(t)x'(t) ~ -Cf(x(b))) for all t E [b,c).
where
then
a(t)x'(t) :::;
Proof. Let x(t) be a solution of equation (4.1.2) satisfying the hypotheses

of the lemma. Since w'(t) + w 2 (t)f'(x(t))/a(t) = -q(t), an integration
for bl :::; t :::; c in view of (4.1.47) gives
(4.1.48)
for b:::; t :::; c. Now, the integral in (4.1.48) is nonnegative, so from the
condition f'(x) ~ 0 for x#-O and the definition of w, we find that
x(t)x'(t) < 0 on [b, c). If x(t) > 0, we let u(t) = -a(t)x'(t), then
(4 .1.48) becomes
u(t)
Cf(x(t))
It
b
f(x(t)) f'(x(s))( -x'(s)) u(s)ds.

j2(x(s))
Define
k(t,s,y)
f(x(t))
f'(x(s))( -x'(s))
j2(x(s))
y
for t and s in [b, c) and y E [0,00) and observe that in this domain
k(t, s, y) is nondecreasing in 11 for fixed t and s. Therefore, Lemma
4.1.1 applies with G(t) = Cf(x(t)), and we have u(t) ~ v(t), where
v(t) satisfies
provided v(s) E [0,00) foreach s E [b,c). Multiplying this equality

by 1/ f(x(t)) and differentiating, we obtain v'(t)/ f(x(t)) == 0, so that
v(t) == v(b) = Cf(x(b)) > 0 for all t E [b, c). Thus, by Lemma 4.1.1,
a(t)x'(t) :::; -Cf(x(b)) for b:::; t < c. The proof for the case when x(t)
is negative follows from a similar argument by taking u(t) = a(t)x'(t) and
G(t) = -Cf(x(t)).
Lemma 4.1.3. Suppose conditions (4.1.9) and (4.1.10) hold, and
lim If(x)1 = 00.
iXi-400
(4.1.49)
If x(t) is a solution of equation (4.1.2), then
oo
a(s) w 2 ( s)1' (x(s) )ds < 00,

lim w(t) = 0,
t-400
(4.1.50)
(4.1.51)
289

and
CXl 1
[00
= [ t a(s) w 2(s)f'(x(s))ds + t q(s)ds
w(t)
(4.1.52)
for all sufficiently large t, where w(t) is as in Lemma 4.1.2.

Proof. Let x(t) be a nonoscillatory solution of equation (4.1.2). If
(4.1.50) does not hold, then it follows from (4.1.10) that there exist t2 >
tl ::;> to and a positive constant C 1 such that inequality (4.1.47) holds with
T = t2, Tl = tl and C = C 1 for all t::;> t2' For the case when x(t) > 0
for t::;> tl it follows from Lemma 4.1.2 and its proof that x'(t) < 0 and
a(t)x'(t) ::; -CI/(X(t2)) for t::;> t2, or x'(t)::; -C1 f(x(t2Ja(t) for
t ::;> t2' Integrating this inequality from t2 to t, we find
x(t) ::; X(t2) - CI!(X(t2)
t2
- ()ds ---+ a s
00
as
t ---+ 00,
which contradicts the fact that x(t) > 0 for t ::;> tl' The prooffor the case
when x(t) < 0 for t::;> tl is similar and hence omitted. This completes
the proof of (4.1.50).
Now, since
- q(t),
we have
w(T) +
[T ats) w2(s)J'(x(sds
= w(t) +
[T q(s)ds,
(4.1.53)
which together with (4.1.10) and (4.1.50) implies that limT---+oo w(T) exists, say, w(T) ---+ C 2 as T ---+ 00, where C 2 is a constant. Then from
(4.1.53), we find
w(t) = C 2 + [ t
CXl
q(s)ds +
[CXl
t
a(s) w 2 (s)J'(x(s)ds,
t ::;> tl'
(4.1.54)
To prove that (4.1.51) and (4.1.52) hold it suffices to show that C 2 = O.

First suppose that x(t) > 0 for t::;> h. If C 2 < 0, then (4.1.10) and
(4.1.50) imply that there exists TI > tl such that
Il
q(s)ds
CXl 1 2 '
::; -C-2 for t::;> Tl and
- ()w (s)f (x(s))ds < - C
-2.
4
n as
4
Now (4.1.54) implies that (4.1.47) holds on m,oo) with b = T I . But,

then by an argument used above, Lemma 4.1.2 and its proof lead to a
290
Chapter 4
contradiction to x(t) > 0 for t 2:

then there exists T2 > tl such that
U'(t)
tl.
On the other hand, if C 2 > 0,
a(t)x'(t) > C 2
f(x(t)) - 2
Therefore, we have
()f'(x(s))(X'(S))2 d
j2(x(s))
s
iTo a s
> C2
2
f'(x(s))x'(s) ds = C 2 In f(x(t))
j2(x(s))
2
f(x(T2))
iTo
The last inequality together with (4.1.49) and (4.1.50) implies that x(t)
is bounded from above. Since x' (t) > 0 for t 2: T2 and f' (x) 2: 0 for
x -I 0, it follows that f(x(t)) 2: f(x(T2 )) for t 2: T 2. Therefore, we
obtain
x'(t) > C2 f(x(t)) > C2 f(x(T2)) for t 2: T2 .
- 2 a(t)
- 2
a(t)
Integrating the above inequality from T2 to t, we get
C f(X(t2))
x(t) 2: x(T2) + ~
2
it
T2
-1()ds,
a s
which contradicts the boundedness of x(t). This completes the proof that
C2 = 0 for the case x(t) > 0, t 2: tl. A similar proof holds for the case
when x(t) < 0, t 2: t l .
We remark that in view of Lemma 4.1.3 an alternative proof of Theorem

4.1.1 without the assumption (4.1.11) can be given as follows: Since f(x)
satisfies (4.1.5) it also satisfies (4.1.49), hence the integral equation (4.1.52)
holds. Let x(t) be a nonoscillatory solution of (4.1.2) which satisfy the
integral equation (4.1.52). Then, we have
x'(t)
1
f(x(t)) 2: a(t)
['XJ
it
q(s)ds,
Integrating the above inequality from

Jxoo du/ f( u), we find
t
itl
x'(s) d
f(x(s)) S
and hence
G(X(tl)) - G(x(t)) 2:
r(t)
du
ix(t,)
f(u)
i 1
t
t,
1
-()
ass
00
to t
and denoting G(x)
G(x(tIl) - G(x(t)),
q(u)duds --+
00
as
t--+oo,
Oscillation theory for superlinca.r differential equations
291

We are now in the position to prove the following result.
Suppose conditions (4.1.5) and (4.1.9) hold,
Theorem 4.1.7.
the essential infimum of J'(x) on any

closed set that excludes zero is positive
and
11 f~~)
Suppose that Q(t)
= 00 = [1
f~~) .
(4.1.56)
- 00,
(4.1.57)
00,
(4.1.58)
= ft q(s)ds exists, and

DO
i
1
lim inf
t--+oo
then either
(4.1.55)
00
to
. to
1
-(-) Q( s )ds >
a s
1 ( + ) 2
_
-(-) Q (s) ds a s
or
lim sup
t--+=
lto
_(1) (Q(s)
a s
roo
ls
_(1) (Q+(u2 dU ) ds
a U
00,
(4.1.59)
where Q+(t) = max{Q(t), O} implies that equation (4.1.2) is oscillatory.

If instead of (4.1.55) condition (4.1.28) holds, then
(4.1.60)
implies that equation (4.1.2) is oscillatory even without assuming the conditions (4.1.56) and (4.1.57).
for t:::: to. Then by Lemma 4.1.3 equation (4.1.52) holds. Thus, we have
x'(t)
f(x(t
:::: a(t) Q(t),
t:::: to
(4.1.61)
This together with (4.1.57) and (4.1.56) implies that x(t) is bounded below
away from zero. Hence, by (4.1.55), J'(x(t:::: c > 0 for some constant
c. From (4.1.61), we find
(
X' ( t ) )
f(x(t
1
+
:::: a2(t) (Q (t
Chapter 4
292
00
(f~~~;)
a(s)
f'(x(sds 2 c
00
ats) (Q+(s2ds.
Now, if condition (4.1.58) holds, then (4.1.52) implies that w(t) -+

-+ 00, which is a contradiction.
00
as
Next, suppose (4.1.59) holds. It is in view of (4.1.57) equivalent to either
. lt
(I) hmsup
t-4oo
to
Q( s) ds -_
-(-)
a s
or (II)
00,
1 1
00
to
-1()
00
ass
-1() (Q + (u
aU
duds -_
00.
If (I) holds, then integrating (4.1.61) and using (4.1.5) we arrive at a

contradiction.
If (II) holds, then since
li~~p 1: ats) ( Q(s) + 1 a(u) (f~~~~)))

00
lim sup
t-400
lt
to
_(1) (Q(s)
it follows from (4.1.52) that
x'(t)
f(x(t
cl
ass
a(t) Q(t)
+ a(t)
00
a(s)
f'(X(UdU) ds
_(1) (Q+(U))2 dU) ds
aU
CXl
(4.1.62)
(X'(S)2,
f(x(s)
f (x(sds.
(4.1.63)
Integrating (4.1.63) and using (4.1.62), we obtain a contradiction to condition (4.1.5).
A similar type of argument leads to the following oscillation criterion

for (4.1.7) with, > 1. It is applicable in the case when condition (4.1.57)
is not satisfied. Clearly, for this equation condition (4.1.28) is also not
satisfied.
Theorem 4.1.8. Suppose Q(t) = ftOO q(s)ds is defined and condition

(4.1.9) holds. Then either
roo
1
(+
2
ito a(s)6(s) Q (s ds
where 6(t) = .ftto dsja(s),
can be very large),
ito a s s
00,
to 2 1
(4.1.64)
or for some constant a> 0 (notice that a
limsup rt_(1) (Q(s)+al

t-400
( / ()(Q+(u))2 dU)dS=00, t2 to21

a U6 U
(4.1.65)
293
implies that equation (4.1.7) with "( > 1 is oscillatory.

Proof. Let x(t) be an eventually positive solution of (4.1.7). Then by
Lemma 4.1.3, (4.1.52) holds. This implies that limHoo a(t)x'(t)Ix"I(t) =
O. Hence, for any E > 0 there is a to such that - E ::; a( t lx' (t) I x'Y (t) ::; E
for t 2' to. Integrating this inequality from to to t, we get
Thus, there exist a large t1 2' to and positive constants 61, 62 such that
- 61~1(t) ::; -x1-'Y(t) ::; 62~dt) for t 2' t 1, or "(X'Y-1(t) 2' 616(t) for
t 2' t 1, where 6 = "(161, Now, since
(x'(s)?
a(sh x'Y +1( s )ds 2'
t.
00
00
(s)) ds,
a ()
s 6 ()(Q
s
we can proceed as in Theorem 4.1. 7.
(4.1.66)
The following corollary is an immediate consequence of Theorem 4.1. 7.

Corollary 4.1.6. Suppose conditions (4.1.5), (4.1.55) and (4.1.56) hold.
If Q(t) = f tOO q(s)ds exists, and
liminf
t--+oo
to
q(s)ds >
( 4.1.67)
00
and either
(4.1.68)
or
lim sup
t-+oo
It
to
[Q(s)
1=
s
(Q+(U)?dU] ds
00,
(4.1.69)

Remark 4.1.4.
1. Conditions (4.1.68) and (4.1.69) can be combined into the following

convenient form
lim sup
t-+oo
It
to
[Q(s)
+ s(Q+(S))2] ds
00.
This follows by changing the order of integration in (4.1.69) and then using
(4.1.67).
2. Butler's theorem [4] is in fact Corollary 4.1.6 without (4.1.56). In the
proof [4, pp. 79] he attempted to show that either (4.1.56) or (4.1.28) must
Chapter 4
294
hold in general, but the proof only works when f' is continuous in a neighborhood of zero, a condition which is not assumed in the hypotheses. In
other words, Corollary 4.1.6 holds if (4.1.56) is replaced by the assumption
tha.t f is continuously differentiable.
3. Under further restrictions on the negative part of Q(t), Butler [4]
showed that (4.1.69) is also necessary for the oscillation of (4.1.1). More
deta.ils of this will be given in Section 4.3.
ft
Corollary 4.1.7. Suppose Q(t) =
OO
1 ~(Q+(s))2ds
00
to
q(s)ds is defined. If either
00,
to:::: 1
(4.1.70)
or for some positive constant 0,

lim sup
t-+oo
it
to
[A(S)
+0
100 ~(Q+(u))2dU]
s
ds
(4.1.71)
00,
then equation (4.1.6) with 'Y> 1 is oscillatory.

Proof. It follows from Theorem 4.1.9 by taking a(t) = 1.
Remark 4.1.5.
1. Either (4.1.70) or (4.1.71) is implied by the condition

lim sup
t-+oo
it
to
[Q(s)
+ 8(Q+(s))2] ds
00.
However, unlike Remark 4.1.3(1), the converse may not be true.

2. We can get a stronger condition from (4.1.71) as follows: Let 81 > 0,
and define
Then the condition

lim sup
t-+oo
it
to
[Q(s)
+ 82100 ~Q~(U)dU] ds =
00,
where 82 > 0 is arbitrary implies oscillation. This procedure can be

iterated to get successively stronger conditions. This can be proved rather
easily by substituting the estimate (obtained from (4.1.52) and (4.1.66)
with a(t) = 1 and 6(t) = t)
x' ((t))
X7 t
Q(t)
+ 81 ~(Q+(s))2ds
00
295
into (4.1.52).
For the formulation of our next theorem, we note that if (4.1.10) holds,
then the function ho(t) = (1/ ya(t)) j;'OO q(s)ds is well defined on [to, (0).
Further, as long as the improper integrals involved converge, we can define
00
hdt)
hn+,(t)
([ho(s)]+)2 ds
r ([h
..
O(')
Ik
~:~:;Jr d"
"
1,2,
where k is an arbitrary positive constant and [ho(t)]+ = max{ho(t), O}.

Theorem 4.1.9. Suppose (4.1.5), (4.1.9), (4.1.10), (4.1.28) are satisfied,
and there exists a positive integer N such that h n exists for n =
0,1, ... , N. If for every constant k > 0,
+ k hN(t) >
ho(t)
y'a(t)
(4.1.72)
for all sufficiently large t, and
oo
_1_ [ho(S)
y'a(t)
+ k hN(S) ds
Va(S)
(4.1.73)
00,

Proof. Suppose equation (4.1.2) has a nonoscillatory solution x(t), and
let Ix(t)1 > 0 for t 2: tl 2: to. Since the hypotheses of Lemma 4.1.3 are
satisfied from (4.1.52), we llld
00
1
(4.1.74)
w(t) 2: vIaJl)ho(t) + k t a(s) w2(s)ds for t 2: tl'
Note that (4.1.50) implies that
00
tJ
- ()w2(s)ds <
a s
(4.1.75)
00.
From (4.1.74), we have w(t) 2: y'a(t)ho(t), which gives

'11.'2 (t)
2: a (t) [( ho (t)) +]2 .
(4.1. 76)
Thus, (4.1.74) leads to the inequality w(t) 2: y'a(t)ho(t) + khN(t) for

t 2: tl' Now, (4.1.72) implies that there exists a t2 2: tl such that
x'(t)/f(x(t)) 2: 0 for t 2: t 2 , and hence
x'(t)
1
[h ()
f(x(t)) 2: Ja(t)
0 t
k hN(t)
+ Ja(t)
for
2: t2'
296
Chapter 4
Integrating the above inequality, we obtain
'X(t)
X(t2)
1t
v, >
-df(v,) -
t2
which contradicts condition (4.l.5).

Example 4.1.5.
xl/(t)
Here a(t)
h ( )
~'
NS
h()
k d
--o
s+
-s
+ 4t7/ 4 [6+3cost+4tsint](x 3 +x) =
O.
(4.l.77)
t>O.
= 1, so (4.l.9) holds. We also have
00
ho(t) =
q(s)ds =
2 + cost
1
t3/4
> t3/4'
which implies that (4.l.72) and (4.1.73) are satisfied with N = l. Observe
that f'(x) = 3x 2 + 1 > 1 for x> 0, so (4.l.28) holds with k = l. Thus,
all the hypotheses of Theorem 4.l.9 are satisfied and hence all solutions of
(4.l. 77) are oscillatory.
4.2. Further Results on

Superlinear Oscillations
Here, we shall establish a more general oscillation criteria for superlinear differential equations under consideration. For this, we introduce the
following four conditions on the function f each of which includes as a
special case the function f(x) = Ixl l sgn x, "( > l.
(F d
is such that
Joo
v;:;:
min
~d ]2
{ [1 1
and
dv, <
00
feu)
inf
x>o
00
00
J- oo
and
v;:;:
dv, <
(4.2.1)
00
v,
du
feu)
'
(F 2 ) Condition (4.2.1) holds, and

min
{ ~r;t Jf'(x) ixrOO .fF(U5

f(v,~ dv"
~~t Jf'(x) ix
oo
.fF(U5
f(v,) dv,
> O.
297
(F3) Condition (4.2.1) holds, f'(x) is increasing on IR+ and decreasing

on IR-, and
min
{~~t f'(x)
1 f~:)'
inf f'(x)
00
x<o
-00
f(U)
> o.
(F4) Condition (4.2.1) holds, f'(x) E C1(IR-UIR+,IR+) with xf"(x):2 0

for x -=f. 0 and f(x)f"(x)/(f'(x2 is bounded on lR - {o}.
We shall show that (F 4),*(F 3 ),*(F 2 ),*(Fd.
(i)
(F 4),*(F 3 ). Suppose (F4) holds and consider a positive constant c
so that f(x)f"(x)/(f'(xW::::: c for all :.c -=f. o. Then for every x -=f. 0, we
have (f1/(x)/(f'(x2) sgn x ::::: (c/ f(x sgn x, and consequently
r(sgn x)oo 1"(u)
ix
r(sgn x)oo du
(f1(u2du ::::: c ix
f(u)

1
-- -
f'(x)
1
l(s gn x)oo du
lim - - ::::: c x
f(u)
lul ..... oo f'(u)
for
x -=f.
o.
(4.2.2)
But, since f is increasing on lR - {O}, 2J:/2du/f(u):2 x/f(x) for

all x -=f. 0, which in view of the fact that f is strongly superlinear gives
limlxl ..... oo(f(x)/x) = 00. This means that limlxl ..... oo f'(x) = 00. Therefore,
(4.2.2) leads to
r(sgn x)oo du
f (x) ix
f(u) > ~
for every
x -=f. 0
and so (F3) is satisfied.

(ii)
(F 3 ),*(F 2 ). This implication is obvious.
(iii) (F 2 ),*(F 1 ). We assume that (F2) is fulfilled. Then there exists a

positive constant k such that
J f'(X)
r(sgn x)oo Jf'(u)
ix
f(u) du > k
for all
x -=f. O.
Therefore, for x -=f. 0, we have

(
r(sgn x)oo
ix
J7'[U5 ) -IF(X5
f(u) du
and hence
lx
(s gn x)oo [l(sgn x)oo
J7'[U5
f(x)
k
sgn x > f(x) sgn x
Jf'(z)
r(sgn
f(u) du 7(z)dz > k ix
X)CXl
du
f(u)
298
Chapter 4
[ r(sgn
ix
J f'(u) dU]2
r(sgn
f(u)
2: k ix
x)oo
x)oo
du
f(u)
for every
0,
which means that (F d holds.

Remark 4.2.1. The case f(x) = lxi' sgn x, x E JR h > 1) is a classical
example of the function f which satisfies all the above conditions. Some
more examples of such functions f are listed in the following:
Example 4.2.1. Consider the function
lxi' [A + sin(ln[l + IxID] sgn x, x E JR

where "f > 1 and A > 1 + (lh). This function f E C(JR, JR) has the
sign property xf(x) > 0 for x # o. Also, f is strongly superlinear, since
f(x) sgn x 2: (A - 1) lxi' for every x =I- O. Moreover, f is continuously
f(x)
differentiable on JR - {O} with
f'(x) =
"flxl,-l [A + sin(ln[l + Ixl])] + ~II cos(1n[l + IxID, x # o.

1+ x
and consequently
f(x) sgn x
Jf'(x)
>
("f [A+ (1 + ~)] Ixl,-lr/2

A-I
("f [A + (1 + ~) ]) 1/2
Ixlh+1)/2
for all
So, f(x)/Jf'(x) is strongly superlinear (i.e. (4.2.1) holds), since

1)/2> 1. Furthermore, for any x # 0, we find
(s g n
x)oo
v'P(U5
f(u) du
_
-
>
O.
h+
1J
00
Ixl
f'(u)
f(u)
--=---..:.~-"-du
("f [A - (1 + ~)]) 1/2 100 u- h +1)/2du

A+ 1
2("f [A - (1 + 1)])1/2
Ixl
--''-----''----'---,---'--'-..:.~-I x I(1-,) /2
(A + l)h - 1)
299

Therefore, for all x -=I- 0, we have
Jf'(x)
(s gn
x)oo
Jf'(u)
21' [,\ - (1 + ~)]
f(u) du >
('\+1)h-1)
> 0,
which means that (F 2 ) holds.

Example 4.2.2. Consider the continuous function f(x) = x + Ixl"! sgn x,
x E JR, where I' > 1. Obviously, xf(x) > 0 for x -=I- o. Also, we
observe that f(x) sgn x ~ Ixl"! for x -=I- 0 and consequently f is strongly
superlinear. Furthermore, f E C 2(JR - {O}, JR) with f'(x) = I'lxl"!-l + 1
and f"(x) = I'h - 1)lxl"!-2 sgn x for x -=I- o. Thus, f'(x) > 0 and
xf"(x) > 0 for x -=I- o. Moreover, f(x)/Jf'(x) is strongly superlinear,
since for Ixl ~ 1, we have
f(x) sgn x
v'f'(x)
>
where
h + 1)/2 > 1.
Ixl"! + X
hlxl"!-l + 1)1/2
Ixl"!
h + 1)1/2
IxlC'Y+1)/2
'
Now, for every x -=I- 0, we find
f(x)f"(x)
(f'(X))2
Clearly, f(x)f"(x)/(f'(x))2 is bounded on JR - {O}, which means that

(F4) holds.
Example 4.2.3. Consider the function f(x) = (lxI 2"! sgn x)/(l + Ixl"!),
x E JR where I' > 1. Clearly, f E C(JR, JR) and xf(x) > 0 for every
x -=I- O. This function is continuously differentiable on JR - {O} and
f is strongly superlinear and for all x
Moreover,
(s gn
x)oo
roo
du =
du
f(u)
ilxl f(u)
roo
-=I- 0,
du
ilxl u 2 ,,!
roo du
+ ilxl
u"!
IxI 1- 2"!
IxI 1-"!
+
-.
21' - 1
I' - 1
Chapter 4
300
Furthermore, we find
yfF(x)
/,1/2Ixl(2,-1)/2 (2
f(x)
sgn x
Ix1
/,1/2 (lxi' + Ixl')

Ixl(2,+l)/2
<
for all x with Ixl 2: 21h. This in view of

validity of (4.2.1). Now, we have
[J
(sgn
x)oo
Jx
(sgn
for x
i= O.
+ Ixl,)1/2
2,
:s;
(2/,)1/2
Ixl(r+l)/2
h + 1)/2 >
1, ensures the
E"Ei du ] 2
f(,,)
x)oo
f(1t)
It is easy to verify that this function satisfies condition (F 1).
We are now in the position to present a more general oscillation criteria

for (4.1.3).
Theorem 4.2.1. Suppose condition (F 1 ) holds and let p(t) E C 1([to, 00),
lR+) be such that
t:(t)
."
to
a(s)p(s)
ds
and
lim c.(t)
t-+oo
00 .
( 4.2.3)
2: to. (4.2.4)
Moreover, let
{3(t)
a(t)p'(t) - p(t)p(t) 2: 0
If
liminf
t-+oo lto
and
1
lim sup t:( )
t-+oo
." t
it
to
and
{3'(t) < 0 for
p(s)q(s)ds > -
( )1 ()
asp s
is
to
(4.2.5)
00
p(u)q(u)duds
00,
(4.2.6)

Proof.
Suppose (4.1.3) possesses a nonoscillatory solution x(t) on
[T,oo), T 2: to. Without loss of generality, we assume that x(t) i= 0
for every t 2: T. We observe that the substitution y = -x transforms
(4.1.3) into the equation
(a(t)y'(t))'
+ p(t)y'(t) + q(t)f*(y(t))
0,
where f*(y) = -f(-y), Y E 1R. Since the function f*(y) is subject to

the same conditions as on f(x), we can restrict our discussion to the case
where the solution x(t) is positive on [T,oo).
Oscillation theory for superlinea.r differential equations
Let w(t) be defined by w(t)

T, we have
= p(t)a.(t)x'(t)/f(x(t)),
I
[(a.(t)XI(t))'
( x'(t)
(t) = p(t)
f(x(t)) - a.(t) f(x(t))
and consequently for every t

I
(t) = - p(t)q(t)
Thus, for any t
t ~ T. Then for
)2 f I(x(t))1+ a.(t)pI(t) f(x(t))

XI(t)
T,
Xl
(t)
+ (3(t) f(x(t))
T, we find
301
t (3(s)x
(s)
rt w 2a.(s)p(s)
(s)f'(x(s))
f(x(s)) ds - .IT
ds.
l
iT p(s)q(s)ds = -w(t) + w(T) + iT
But by the Bonnet theorem, for a fixed t ~ T and for some

we have
XI(S)
JT (3(s) f(x(s)) ds
(4.2.7)
- a.(t)p(t) 11' (t)f (x(t)).
(3(T)
XI(S)
f(x(s)) ds = (3(T)
JT
r(ry)
Jx(T)
7) E
[T, t],
du
f(u)
and hence, since (3(T) 2: 0 and
du
{
f(u) <
r(ry)
Jx(T)
we obtain
t p(s)q(s)ds
JT
l(T)
(Xl
t (3(s) f(x(s))
x/(s)
ds
JT
where K = (3(T)
for t ~ T that
if
Ix'7T) du/ f(u).

:s; - w(t)
x(7)) < x(T)

du
f(u) if x(7))
:s; K
for all
~ x(T),
t ~ T,
(4.2.8)
Therefore, in view of (4.2.8), we conclude
+ w(T) + K
t w 2a.(s)p(s)
(s)f'(x(s))
ds.
iT
(4.2.9)
Now, with respect to the integral
I(T, t) =
JTt
a.(s)p(s) w2(s)j'(x(s))ds
there are two cases to consider:

Case l. I(T, (0) is finite. In this case there exists a positive constant N
such that
(4.2.10)
I(T, t) :s; N for t ~ T.
302
Chapter 4
Furthermore, by using the Schwartz inequality, for t
t
I
x'(s)Jf'(x(s)) dsl2
iT
f(x(s))
Ja(s)p(s)
[It a(s)~(s)dsJ
<
Thus, in view of (4.2.10), we find
IiTt
T we obtain
2
w(s)Jf'(x(s)) ] d 1
Ja(s)p(s)
s
J(T,t).
1
x'es)
12 < N t
ds
f(x(s) J f'(x(s))ds
T a(s)p(s)
N~(t)
for all t
T.
(4.2.11)
Next, using (FI)' we get
l(t)
CXl
du
< !vI
feu) -
l(t)
CXl
Jf'(u) ]
du
feu)
(4.2.12)
t ~T
where M is a positive constant. Now, setting
KI =
d
feU) > 0 and
x(T) U
CXl
and using (4.2.12) for every t
w(s)
l i t x'es) 1
IJTt a(s)p(s)
ds = iT f(x(s ds =
IKI
-1:) f~:)
< KI + M
[l
CXl..jF(Uj
x(t)
Kl
feu) du
]2
00
x(T)
J f'(U)
f( ) du > 0
U
we obtain
T,
K2 =
(xU) du
ix(T) feu)
1:) f~:)
[ l (t)..jF(Uj
X
Kl + M K2 -
x(T)
feu) du
]2
r*)..jF(Uj 1]2
< Kl + M [K2 + 1iX(T) feu) du
KI
+M
[K2
+ Ilt
f~~~s;)
Thus, by (4.2.11) for every t
I i a(s)lp(s) W(S)dSI
f'(X(S)dsIJ
T, we have
< Kl + M [K2 + (N~(t1/2r

(Kl
+ MK?) + 2MK2v'Ne/2(t) + MN~(t).

By condition (4.2.3), we can choose a To
rTO
ito a(s)p(s)
303
> to so that
ds ~ l.
(4.2.13)
and hence, in view of (4.2.13), we get
Ilt
a(s)1p(s) w(s)dsl :S
where T* = max {To , T} and C

from (4.2.9) it follows that
C~(t)
for every
~ T*,
= Kl +MKi +2MK2ffi +MN. Now,
lt p(s)q(s)ds :S - w(t)
+ w(T) + K
for
and thus, by taking into account (4.2.14), we obtain for t

(
(4.2.14)
~T
~
T*,
iT a(s)p(s) iT p(u)q(u)duds
rt
< - iT a(s)p(s),w(s)ds + [w(T) + K] iT a(s)p(s) ds

t
< Ilt a(s)1p(s) w(S)dSI + Jw(T) + KJl a(s)1p(S) ds :S

where C* = C
( ) ()
toasps
=
+ Jw(T) + KJ.
is
to
C*~(t),
Therefore, for every t ~ T*, we have
p(u)q(u)duds
r
+ [ rT P(U)q(U)dU] t _(
ito
iT asp
t
r
+ iT a(s)p(s) iT p(u)q(u)duds
r
T
()1 ( )
p(s)q(u)duds
ito asp s ito
)1 ( )ds
s
:S
11:
a(s{p(s)
1:
P(U)q(U)dUdSI +
11:
P(U)q(U)dUI
+ C*~(t)
a(s)1p(s) ds
and consequently, in view of (4.2.13) it follows that
to
( ) ()
asp s
is
to
p(u)q(u)duds:S Cl~(t)
for all
t ~ T*,
304
Chapter 4
where
This contradicts condition (4.2.6).

Case 2. I(T, 00) = 00. In view of condition (4.2.5), from (4.2.9) it follows
that for some constant m,
-w(t) ?: m
+ I(T, t)
for every
t?: T.
(4.2.15)
Choose a TI ?: T such that M = m+I(T, Td > O. Then, (4.2.15) ensures

that w(t) is negative on [TI' 00). Now, (4.2.15) gives
~W2(t)f'(X(t))
>
m+I(T,t)
x'(t)f'(x(t))
f(x(t))
and consequently for all t?: T I ,

1 m
n
+ I(T, t) >
1 f(x(TI))
n f(x(t)) .
Hence, m+I(T,t) ?: Mf(x(TI))/f(x(t)) for t?: T I . Thus, (4.2.15) yields

x'(t) :S -MIf(a(t)p(t)) for every t?: T I , where MI = Mf(x(TI )) > O.
Thus, we have
x(t) :S x(Td -
MI
()ds
asp s
( )
T,
for all
t?: T I ,
which because of condition (4.2.3) leads to the contradiction that

=
-00.
lim x(t)
t ..... oo
The following corollaries, which are well known basic results in the literature, are immediate from Theorem 4.2.1.
Corollary 4.2.1. If
liminf
t ..... oo
and
11 l
lim sup t--+oo
t q(s)ds
lto
to
to
> -
q(r)drds
(4.2.16)
00
00,
(4.2.17)

The following result extends Corollary 4.2.1 to equation (4.1.1).
305
Corollary 4.2.2. If conditions (FI), (4.2.16) and (4.2.17) are satisfied,

A generalized criterion for the oscillation of (4.1.1) is the following result.
Corollary 4.2.3. Suppose condition (FI) holds and let p(t) E Cl([to, (0),
JR+) be such that p' (t) is nonnegative and decreasing on [to, (0). If
condition (4.2.5) holds, and
lim sup
t-+oo
[I
to
-1()ds]
p
-ll l
t
to
1
-()
PS
to
p(u)q(u)duds
00,
(4.2.18)

Remark 4.2.2. Since the function p(t) in Corollary 4.2.3 is positive
on [to, (0) and p'(t) is nonnegative and bounded above on [to, (0), it
follows that p(t):-:; /1t for all large t, where /1 > 0 is a constant. This
ensures that
= 1
(4.2.19)
- ()ds = 00.
to P s
Next, let r be the class of functions defined as follows: p(t) E r if

and only if p(t) E CI([to, (0), JR+) such that p'(t) is nonnegative and
decreasing on [to, (0). This class is large enough, in fact p(t) Erin each
of the following cases:
(i)
(ii)
p(t) = tA, t ~ to for A E [0,1].

p(t)=ln>'t, t~to for A>O, where to >max{l,e>'-l}.
(iii) p(t) = e'ln t, t ~ to for A E (0,1), where

to > max { 1, exp
C~
A-
~) } .
(iv) p(t)=t/lnt, t~to where to~e2,

(v) p(t) = t 1 / 2 [5 + sin(ln t)], t ~ to
and the simplest case
(vi) p(t) = 1 for t ~ to.
Now, if p(t) = t>', t ~ to for A E [0,1] in Corollary 4.2.3, then we
obtain the following interesting result.
Corollary 4.2.4.
Suppose condition (FI) holds.
oscillatory if for some A E [0,1],
liminf
t-+= Jto
s>'q(s)ds >
00
Equation (4.1.1) is
Chapter 4
306
and
limsuptA-l
t-tCXJ
lim sup - 1
t-too In t
Example 4.2.4.
(0x'(t)),+
t S-A ltot
lto
it I1
-
to S
uq(u)duds
to
U A q(1L)duds
for 0 <:::;)..
00
<1
for A = 1.
00
l~x'(t)+ l-sint+~(2+cost)jf(x(t))
2t
= 0,
2y t
(4.2.20)
t ::;> to = 7r /2
where the function f is as in Examples 4.2.1 - 4.2.3. Let pet) = 0
t::;> to. Then, we have f3(t) = a(t)p'(t) - p(t)p(t) = 0 and
a(t)p(t) = t,
~(t)
(p(s)q(s)ds
1t o
and
11r/2
11r/2
1
t
to
1 1/
to
[-.JSSins+
d[y'S(2+coss)]
I ( / ) 1t n 2t 7r
to S
( ) ( ) ds = In -to -+
asp s
as
00
t -+
for
00,
1;-;,(2+COSS)j ds
2ys
0(2+cost)-2V7r/2 ::;> 0-2V7r/2
y'uq( u)duds
([1v's--;Y"2
::;>
2Fj
In(2t/7r)11r/2
ds
20
2,Ji/2
In(2t/7r) - In(2t/7r) - 2y
"2
and consequently,
liminf1t y1Sq(s)ds >
t-too
and
lim sup I (21 / )
t-too
n t 7r
to
it I1
-
to S
to
00
y'uq(u)duds
00.
Now, Theorem 4.2.1 ensures that (4.2.20) is oscillatory.

" [1
x (t)+ -
~tsint+2t ~t(2+cost)
y~
y~
j [x(t)['Ysgnx(t)
0,
t ::;> to = 7r/2
(4.2.21)

where , > 1 is a constant. For every t
and
1lt 11"
-1-
to S
nt
to
>
uq(u)duds
307
to, we find
1 it/2 [1JS - -;2yFl2" J ds
In t
7r
F[In2"-l.
Vi
2
2 1nt -2 Y 2"+lntY2"
7r
Thus, all conditions of Corollary 4.3.4 are satisfied and hence (4.2.21) is
oscillatory. On the other hand, Corollary 4.2.1 is not applicable for (4.2.21).
Indeed, for t ~ to, we have
it [- ~sins+
tq (s)ds =
it
to
<
-
7r/2
7r /2
- Vi
it
7r/2
Thus, for every t
11tl"
to
1r;;(2+COSS)]dS
2sy S
(- 1
- sin s + -3 -1-) ds = -1 cos t
JS
2 sJS
Vi
< - 1 cos t + 2
Y S
to
1
--ds
sJS
= -1
Vi
it (- -
7r/2
cos t - 4 [1
.- -
Vi
1 cos s + -3-) ds
2sJS
2sJS
~j
7r
y2
< - -3 + 4 -.
Vi
7r
to,
q(u)duds :::;
-~+~
fi+4 II
Vi tY2" y-;
~it
[-~+4
IIjds
t
JS
y-;
7r/2
and hence
11tl"
lim sup t-+oo
to
to
q(u)duds <
00,
i.e., Corollary 4.2.1 fails to apply to equation (4.2.21).

Theorem 4.2.2. Suppose (F 2 ) holds and let the functions p(t), (3(t)
and ~(t) be as in Theorem 4.2.1, and conditions (4.2.3) and (4.2.4) are
satisfied. If (4.2.5) holds and for every positive constant k,
limsup :
t-+oo
lt
to
(t - s)n-2
[(t _ S)2p(S)q(s) _
(k )] ds
4 v s
00
(4.2.22)
for some integer n > 1, where v(t) = l/(a(t)p(t)~(t)), t ~ to

then
Proof. Let x(t) be a nonoscillatory solution of equation (4.1.3). Without

loss of generality, it can be assumed that x(t) i= 0 for t ~ T ~ to.
Chapter 4
308
Furthermore, it is enough to consider the case when x(t) > 0 for t 2

= p(t)a(t)x'(t)/ f(x(t)), t 2 T. As in Theorem 4.2.1,
we obtain (4.2.7) and conclude that (4.2.8) and (4.2.9) are satisfied with
K = (3(T) Ix7T) du/ f( u). We consider the following two cases:
T. Define w(t)
Case 1. I(T,oo) < 00. Then, (4.2.10) holds where N is a positive

constant. Furthermore, as in Theorem 4.2.1 inequality (4.2.11) follows.
Condition (F2) ensures that
~ du
J f'(X(t)) (')0
.ix(t)
2 A for t 2 T,
where A is a positive constant. Now, let B

Then by (4.2.23), we obtain for t 2 T,
f'(x(t))
[1
A2
00
x(t)
A2 [B
= Ix7T) Jf'(u)/f(u)du > O.
y7'(Uj duj-2 = A2 [B
f(u)
A2 [B-lt
rt
(4.2.23)
_l
X
(t)
x(T)
y7'(Uj duj-2
f(u)
f~~~~))Jf'(X(S))dSr2
+ 1.iT f~x(:)) J
I(
f'(x(s))ds
I] -2
Using (4.2.11) in the above inequality, we get f'(x(t 2 A2[B+JN~(t)J-2

for t 2 T. As in Theorem 4.2.1 there exists a To 2 to such that (4.2.13)
holds and consequently, we find
f'(x(t) 2
for all
c/~(t)
where c = A2[B + v'NJ- 2

I(
P( t ) q( t ) < - w t
+ (3
2 T*
(4.2.24)
ma.x{To, T},
> O. In view of (4.2.24) equation (4.2.7) gives
()
x' (t)
t f(x(t
2( )
- a(t)p(t)~(t) w t
for all
2T ,
or
p(t)q(t) ~ - w'(t)
Thus, for t 2 T*,
rt
x'(t)
+ (3(t) f(x(t
rt
- cv(t)w 2(t)
for all
t 2 T*. (4.2.25)
rt
X'(S)
.irY - stp(s)q(s)ds ~ - .iT 5t - s)nw'(s)ds + .iT,(t - s)n{3(s) f(x(s ds
- .iT'r c(t - s)nv(s)w2(s)ds .
(4.2.26)
309

But because of (4.2.8), we ha.ve
t Y - s)n-l [fT,(3(u)
t f(x(u))
x'(u)]
du ds
n
x'(s)
fTY - s) (3(s) f(x(s)) ds
n fT
:::;
Kn {t (t _ s)n-1ds
fT'
K(t _ T*)n.
Hence, for every t 2': T*, we obta.in
fT'
(t - s)np(s)q(s)ds
< - {t (t _ s)nu/(s)ds + K(t _ Tt _
h-
+ KJ
(t - T*)n[w(T*)
- t
iT-
iT-
fT-
c(t - stv(s)w2(s)ds
+ KJ + -n
(t - T*)n[w(T*)
_t
- n
4c
[JC(t _ s)nv(s)w(s)
h-
c(t - s)nv(s)w 2(s)ds
(t - s)n-1w(s)ds
!,t (t - s)n-2_(-)ds
1
T'
n(t - s)n-l
2Jc(t - s)nv(s)
]2 ds
1
< (t - T*)n[w(T*) + KJ + -n21t (t - s)n-2_(
)ds.
4c
T'
Now, since
n1
t
it
to
<
<
(t - s)np(s)q(s)ds
~ {T" (t _ s)np(s)lq(s)lds + ~ {t (t _ stp(s)q(s)ds

t fto
t fT"
( t)~
1-
+~
tn
(n2)
4c
(* p(s)lq(s)lds + ( 1 - tT* ) n [w(T*) + KJ
fto
fT"
(t _ s)n-2_1_ds
v(s)
for all
2': T*
(4.2.27)
it follows that
limsup
{t
t-+oo t fto
[(t _ s)np(s)q(s) _
<
T-
fto
n 2 (t _ s)n-2_(1)] ds
4c
v S
p(s)lq(s)lds + w(T*) + K <
00,
Chapter 4
310
Case 2. J(T,oo) = 00. In this case exactly as in Theorem 4.2.1 we arrive

at the contradiction limt-HXl x(t) = -00.
The following corollaries are immediate.

Corollary 4.2.5. In Theorem 4.2.2 the condition (4.2.22) can be replaced
by
1
li~~p t 2
and for some integer n > 1

lim sup -1
t-+oo t n
.ltot a(s)p(s)~(s)ds
it
to
<
(4.2.28)
00
(t - s)np(s)q(s)ds
(4.2.29)
00
Corollary 4.2.6. If condition (4.2.16) holds and
it
1
lim sup -;;(t - s)nq(s)ds
t-+oo t to
(4.2.30)
00
n> I, then equation (4.1.6) is oscillatory.
for some integer
Corollary 4.2.7. If conditions (F 2 ), (4.2.16) and (4.2.30) are satisfied,

Example 4.2.6.
(rix'(t)), +
~rx'(t) +
[-r! sint +
~t-%(2 + cost)]
for
f(x(t)) = 0
to
= 7r/2,
(4.2.31)
where f(x) is as in Examples 4.2.1- 4.2.3. Let p(t) = t 5 / 6 . Then, we have
j3(t) = a(t)p'(t) - p(t)p(t) = 1/(6t), t ~ to and l/v(t) = a(t)p(t)~(t) =
t - to, t ~ to also for all t ~ to,
rtp(s)q(s)ds =
to
1,,/2
t s5/6[-s-1/3sins+~s-4/3(2+COSS)]dS
J"/2
[-v'sSins+ 2 \;;(2+COSS)]dS
Y s
vt(2 + cost) - 2f%.

and hence liminft-+oo
rt d[v's(2
1,,/2
+ coss)]
vt - 2f%.
It: S5/6 q(s)ds
> -
00.
Furthermore, we have for
311
all t;::: to and every constant c > 0,

1
2"
it [
to
(t - s) 2 s 5/6 q( s) - - C ] ds
v(s)
t~
>
(1:(t-s)
(it
[1:
(t-s)
1f/2
u 5 / 6q(U)dU] ds_c(t-2t o)2)
[,;8-2 VEl'2 J
~Vt _ 2 E + ~ (~)3/2 ~
V'2
15
t2
dS_C(t-7r/2)2)
2
_.: (1 _ ~)2
2t
and consequently,
lim sup
t400
~
t
ito
[(t _ s)2 s5/6 q(s) - _(c)] ds =
00.
V S
Thus, condition (4.2.22) holds for every constant c> 0 and n = 2. Now,
Theorem 4.2.2 ensures that (4.2.31) is oscillatory.
We note that Corollary 4.2.5 is also applicable to equation (4.2.31).
x"(t)
[_C 1/ 3 sin t
+ ~C4/3(2 + cost)]
Ix(t)I" sgn x(t)

for
t;::: to
7r/2,
(4.2.32)
where 1 > 1 is a constant. As in Example 4.2.6 equation (4.2.32) is

oscillatory by Corollary 4.2.5 for p(t) = t 5 / 6 and n = 2.
On the other hand, condition (4.2.30) fails and consequently Corollary
4.2.6 cannot be applied to this equation. In fact, for every t;::: to, we have
ito
q(s)ds
i
i
[_S-1/3 sin s
n/2
<
+ ~S-4/3(2 + cos S)]

2
(_s-1/3 sins +
1f/2
< C 1/ 3 cos t
C 1 / 3 cost
~s-4/3) ds
2
it [~S-4/3
+ -11it
t- 1/ 3 cos t +
n /2
11
-"2
ds
cos S
+ ~S-4/3]
2
ds
S-4/3ds
7[/2
[C
1/ 3 -
(7r)-1/3]
'2
<
121
(7r'2)-1/3
Chapter 4
312
Now, if n is an integer with n > 1, then for all t 2: to, we find
~ i t (t -
to
s)nq(s)ds = : i t (t - s)n-l [is q(U)dU] ds

t to
to
< ~ (~)-1/3 (~) i t (t _ st-1ds = ~ (~)-1/3

2 2
tn
/2
2 2
7r
This gives limsuPHoo(l!t n ) J/o(t-s)nq(s)ds <

is not satisfied.
00,
(1 _ ~)
.
2t
i.e., condition (4.2.30)
In Theorem 4.2.2 if the condition (4.2.4) on the function (3(t)

then we have the following result.
fails,
Theorem 4.2.3. Suppose (F2) holds and let the functions p(t), (3(t), f,(t)
and v(t) be as in Theorems 4.2.1 and 4.2.2 such that condition (4.2.3) is
satisfied. If condition (4.2.5) holds and for every constant c> 0,
~ i\t-s)n-2 f(t-s)2 p(s)q(s) - ~(

) (t-s) r~s~ ) 4v s \
asp s
limsup
t-+oo
to
n)2] ds
=
00
(4.2.33)
for some integer n> 1, then equation (4.1.3) is oscillatory.

x(t) > 0 for t 2: T 2: to. Proceeding as in Theorem 4.2.2 - Case 1, we get
(4.2.26) which for t 2: T* takes the form
Proof.
t (t iT'
s)np(s)q(s)ds
< (t - T*)nw(T*) +
- iTo
t c(t ft
(t - s)n-l [(t - s)
arsi;~s) -
n] w(s)ds
s)nv(s)w 2 (s)ds
(t - T*tw(T*)
- iT'
t (t _ s)n-2 [(t _ s) a(s)p(s)

(3(s) _ n]
iT'
Jc(t - s)nv(s)w(s) -
_c_ ds
4v(s)
(t-s)n-l(t-S)~)l2
a(s)p(s)
< (t-T*tw(T*) + .(t_s)n-2
2Jc(t _ s)nv(s)
[(t-s)aCi;~s)
ds
-nf 4v~s)ds.
313
The following corollary is immediate.

Corollary 4.2.8. In Theorem 4.2.3 condition (4.2.33) can be replaced by
it
.
1
lnllsup-;;
t--+oo
to
(t-s)
n-2 [
(3( s)
]2 1
(t-s) () () -n - ()ds < 00
asp s
v s
(4.2.34)
and (4.2.29) holds for some integer n> 1.

(t 1 / 6 X'(t))' +
6~x'(t)+ [-C1/3sint+~C4/3(2+Cost)]
f(x(t)) = 0
to = 1f/2,
(4.2.35)
where f is as in Example 4.2.6. Let p(t) = t 5 / 6 , t ~ 7r /2. Then, we
have (3(t) = a(t)p'(t) - p(t)p(t) = (5 - C 1 / 6 ) /6 > 0 for t ~ 1f/2 and
for
(3'(t) = C
/36 ~ 0 for t ~ 1f/2. Also, a(t)p(t) = t, ~(t)
7/ 6
J:/2ds/s
In(2t/1f) and 1/v(t)=tln(2t/1f), t~1f/2. Now for all t~to,
-1
t2
it [
to
t-s
1
="2
t
Tr/2
(3(s) -2
a(s)p(s)
[
]2 -1d s
1
6s
v(s)
-(t-s)(5-s
2s
/)-2 ] 2 sln-ds
1f
> 6ln 2t ,
-1 6'
1f
(4.2.36)
where b> 0 is a constant. As in Example 4.2.7, we find
~
t2
it
S5/6 q (s)ds
Tr /2
>
-
!i.,;t _ 2 E + ~ (~)3/2 ~.
15
V"2
t2
(4.2.37)
Now, from (4.2.36) and (4.2.37) it follows that the conditions of Theorem
4.2.3 are satisfied, and hence equation (4.2.35) is oscillatory.
Next, we present the following oscillation criteria for (4.1.3) when condition (4.2.22) fails for some integer n > 1.
Theorem 4.2.4.
Suppose condition (F 2 ) holds and let the functions

and v(t) be as in Theorem 4.2.2 such that (4.2.3) (4.2.5) and (4.2.28) are satisfied. If there exist an integer n > 1 and a
function O(t) E C([to, 00), JR) such that for any constant c> 0,
p(t), (3(t) ,
~(t)
lim sup ~ ((t_s)n-2 [(t _ S)2p(S)q(s) _ n c _(1)] ds

t--+oo t n
4 v s
iT
for every T
O(T) (4.2.38)
to, and
00,
(4.2.39)
Chapter 4
314
where n+(t)
= max{n(t), O},

x(t) > 0 for t::::: T ::::: to. We define the function wet) as in Theorem
Proof.
4.2.1 and consider the two cases:
Case 1. l(T, 00) < 00. Proceeding as in Theorem 4.2.2 - Case 1, we obtain
(4.2.26). The rest of the proof is exactly the same as that of Theorem 2.5.12
for H(t, s) = (t - s)n, n> 1 and hence omitted.
Case 2. l(T,oo) = 00. The proof is similar to that of Theorem 4.2.1 Case 2.
Theorem 4.2.5.
Suppose condition (F2) holds and let the functions

and vet) be as in Theorem 4.2.2 such that (4.2.3) (4.2.5) are satisfied. If there exists an integer n > 1 such that
pet), (3(t),
~(t)
liminf -1
t .... oo
tn
it
to
(t - s)np(s)q(s)ds < 00
and there exists a function net) E C([to, 00), lR)

(4.2.39) holds, and for any constant c > 0,
T::::: to,
for every
(4.2.40)
such that condition
Proof. The proof is similar to that of Theorem 4.2.4 except that now we
employ Theorem 2.5.13 instead of Theorem 2.5.12.
In Theorems 4.2.4 and 4.2.5 if condition (4.2.4) fails, then we have the
following results.
Theorem 4.2.6.
Suppose condition (F2) holds and let the functions

and vet) be as in Theorem 4.2.2 such that conditions
(4.2.3) and (4.2.5) are satisfied. If there exist an integer n > 1, and a
function net) E C([to,oo),lR) such that conditions (4.2.34) and (4.2.39)
hold, and for every constant c > 0,
pet), (3(t),
li~~p t n
~(t)
c (
(3( s)
iTrt(t_s)n-2 [(t-s)2 p(s)q(s)- 4v(s)
(t-s) a(s)p(s) -
2] ds
::::: neT)
for every
T::::: to,
Theorem 4.2.7.
pet), (3(t) ,
~(t)

Suppose condition (F 2 ) holds and let the functions
and vet) be as in Theorem 4.2.2 such that conditions
315
(4.2.3) and (4.2.5) are satisfied. If there exist an integer n > 1, and a
function O(t) E C([to, 00), JR) such that conditions (4.2.39) and (4.2.40)
hold, and for every constant c > 0,
1ft(t-st- 2 f(t-S)2p(S)q(s) -
liminf - n
t-+oo t
c ( (t-s) (j3(s)
-(-)
) ( ) - n
4v s
asp s
)2] ds
;::: O(T)
for every T;::: to, then equation (4.1.3) is oscillatory.
Example 4.2.9. Consider the differential equatioll
+ (cost)f(x(t))
(e-tx'(t))'
t;::: to>
0,
(4.2.41 )
f is one of the functiolls defined in Examples 4.2.1 -- 4.2.3. Taking
where
p(t)=l and n=2, weget

v(t)
)]-l
[1- e-(t-t o
1
liminf2"
t
t-+oo
and
to
t > to> 0,
(t-s)2cos s ds
liminf
t-+oo
j.t cos sds
> - 00,
to
= - sin to < 00,
1ft [(t-s)2coss-c(1-e-(t-to))]ds;:::
liminf2"
t-+oo
1t
for
l'
-sinT-K
for every T;::: to, where c and K are positive constants and K is
sufficiently small. Let O(T) = - sin T - K. Next, we consider an integer
N such that (2N + 1)71" + (71"/4) > to. Then for all integers n;::: Nand
(2n + 1)71" + (71"/4) ::; T::; 2(n + 1)71" - (71"/4), O(T) = - sinT - K ;::: 5T,
where 5 is a small constant. Thus, we have
lim
t-+oo
1
t
v(s)(O+(s))2ds;:::
L
00
52
n=N
to
2 (n+l)-(n/4)
s2ds
00.
(2n+1)n+(n/4)
Hence, all conditions of Theorem 4.2.5 are satisfied and therefore equation
Theorem 4.2.8. Assume that condition (4.1.5) holds,
min{~~~f'(X)
1= f~~)' ~~~f'(x) 1-= f~~)}
> 1
( 4.2.42)
and suppose there exists p(t) E C1([to, 00), JR+) such that condition (4.2.3)
holds and a(t)p'(t)
= p(t)p(t), i.e., p(t) = exp (t(p(s)/a(s))ds). If

lim inf F(t) >
t-+oo
- 00
(4.2.43)
Chapter 4
316
and
lim F(t)
t-'>oo
where
F(t)
1
-()
~ t
it
to
does not exist,
1
8
( )1 ( )
asp s
to
(4.2.44)
p(u)q(u)duds,

We note that conditions (4.2.43) and (4.2.44) can be replaced by
-00 < liminf F(t) < limsupF(t) :::; 00.

t-'>oo
(4.2.45)
t-'>oo

x(t) > 0 for t:::- to. We define w(t) = Jx'7t) du/ f(u) for t:::- to. Then,
we have
(a(t)p(t)w'(t))' = p(t)q(t)+a(t)p(t)(w'(t))2 j'(x(t)) for
t:::- to.
(4.2.46)
Integrating (4.2.46) twice from to to t, we find
w' (t)
Cl
( )t+a ()
a (t )p
t p(t )
where
J(to, t) =
and
Cl
= a(to)p(to)w'(to),
w(t)
~(t)
w(t o )
1:
~ t
it
to
to
p(s)q(s)ds + asp
() (s /(t o, t),
a(s)p(s)(w'(s))2 j'(x(s))ds
and
1
~(t) + Cl + ~(t)
1
+-()
it
it
to
1
8
a(s)p(s)
to
1
-()-() J(to, s)ds.
p(u)q(u)duds
(4.2.47)
asp s

mutually exclusive cases:
J(to, t)
we need to distinguish two
Case 1. J(to,oo) < 00. First, we will show that
lim w(t) =
Hoo ~(t)
Let
o.
(4.2.48)
E> 0 be an arbitrary number. We can choose a tl :::- to such that

J(tl' 00) :::;
4.
(4.2.49)
317
Furthermore, using the Schwartz inequality for t 2: t1, we have
where
G(tJ, t)
itl
a(s)p(s)f'(x(s)) ds
and so, in view of (4.2.49), we get
(4.2.50)
If G(t1' (0) <
is bounded on
that G(tJ, (0)
then from (4.2.50) it follows that the function w(t)

and hence (4.2.48) is satisfied. So, we assume
= 00. Then there exists a number t2 > t] such that
w(t])::; (y'Ej2)[G(t],i)j1/2 for i 2: t2, and consequently (4.2.50) gives
00,
[t1' (0)
(4.2.51 )
But condition (4.2.42) guarantees that
f'(x(t))w(t) = f'(x(i))
r= jd(U)
ix(t)
>
for
2: il'
Thus, from (4.2.51) it follows that
w(t)
<
y'E
a(t)p(t) - a(t)p(t)
or
[i
[titl
w(s) d]
a(s)p(s) s
]/2
w(s)
] -1/2 w(t)
a(s)p(s) ds
a(t)p(t)::;
t,
VE
for all
t::::: t2,
for all
(4.2.52)
t 2: t 2
Integrating the above inequality from t2 to t wIth t 2: t 2 , we get
1,
t
w(s) d] 1/2 2
a(s)p(s) s
-
<
VE [~(t)
[lt za(s)p(s)
w(s) d]
s
1/2
t,
-l:z a(s)~(s)dS]
<
VE ~(t).
Therefore, on setting
t3 =
max {t2'
~
y'E
[ltz a(s)p(s)
w(s) dS]
t,
1/2}
Chapter 4
318
for all t
t3, we obtain
[i
()
WS
a(s)p(s)
t,
] 1/2
< .jE ~(t).
Hence, (4.2.52) gives u(t) < E~(t) for all t

the proof of (4.2.48) is complete.
t3. As E > 0 is arbitrary,
Now using (4.2.48) in (4.2.47), we find
it
lim 1:(1)
t-+oo <,
to
- C1 -
()1 ( )
asp s
1
8
1
lim - ()
t-+oo ~
p(u)q(u)duds
it
to
to
( ) ( /(ta,s)ds,
asp
s

Case 2. J(ta, (0) =
(4.2.42), we have c
claim that
It:
00.
>
Define c = infx>a f'(x) Jxoo du/ feu). In view of

1. We can choose {L such that 1 < {L < c. We
lim sup [(l/{L)B(t) - wet)] > 0,
(4.2.53)
t-+oo
where B(t) =
J(ta, s)/(a(s)p(sds. Assume that (4.2.53) does not
hold, then there exists a T1 ~ ta such that
B(t)
{Lw(t)
for
t ~ T 1.
(4.2.54)
It is easy to see that B(t) -+
00, a(t)p(t)B'(t) -+ 00 as t -+ 00 and

(a(t)p(t)B'(t' = a(t)p(t)(w'(tW f'(x(t. Using (4.2.54) for t ~ T 1, we
find the estimate
( a(t)p(t)B'(t') 1/2 (B'(t)1/2

B(t)
a(t)p(t)B'(t)
where {L1
(W'(t2f'(X(t)1/2 >
w'(t)
B(t)
- {L1 wet)
(4.2.55)
= J(c/{L) > 1.
Note that B(ta) = 0, so we can choose B(T2 ) = 1 for some T2 ~ T 1.

Now, integrating (4.2.55) from T2 to t and applying Schwartz's inequality
to the left side of (4.2.55), we obtain
a(t)p(t)B'(t) )) 1/2
( In ( a(T
(lnB(t
2 )p(T2 )B'(T2)
wet)
1/2
~ ..,fiilln w(T2 )'
(4.2.56)
Using (4.2.54) we can estimate the right side of (4.2.56) as follows
wet)
..,fiilln w(T2)
= ..,fiil [In wet) -In w(T2)]

~
..,fiil[lnB(t) -lnw(T2)]
..,fiillnB(t) for t
T 2.
Oscilla,tion theory for superlinea,r differential equations

Thus, In(ka(t)p(t)B'(t))?: /1llnB(t), where k
0, and hence
B-ILI (t)B'(t)
>!
-
319
= 1/(a(T2 )p(T2 )B'(T2 )) >
(4.2.57)
k a(t)p(t)
Integrating (4.2.57) from T2 to t and using condition (4.2.3), we get

00
lIlt
> - - [B l -IL1(T2 )
/11 - 1
Bl-~'l(t)l ?: -k
,
T2
( )1 ( )ds
asp s
---t
00
as t ---t
00,
which is a contradiction. Thus, the assertion (4.2.53) holds, and there exists
a sequence {tn}~=l with limn--+oo tn = 00 and such that
(4.2.58)
Using (4.2.58) in (4.2.47), we find for sufficiently large tn,
( -1
1)
+ P, ~(tn) B(tn)
tn a(s)p(s)
1
r
lto p(u)q(u)duds.
+ ~(tn) + ~(tn) lto
B(t)/~(t) ---t 00
l.t
liminf -1()
~
Cl
(4.2.59)
Since 1//1 < 1 and
t--+=
w(to)
?:
( )1 ()
asp s
to
as t ---t
1
8
to
(4.1.59) implies
00,
p(u)q(u)duds
- 00,
When p(t) == 0 in Theorem 4.2.8 then clearly, p(t) == 1. Thus, we

have the following corollary.
Corollary 4.2.9.
6(t)
It:
Assume that conditions (4.1.5) and (4.2.42) hold and

ds/a(s)---t 00 as t ---t 00. If
1
liminf -(-)
t--+=
and
1t 1
1t 1
to
1
lim -(-)
t--+=
6 t
1
-()
a s
to
to
1
-()
a s
q(u)duds > -
to
00
(4.2.60)
q(u)duds
does not exist, then equation (4.1.2) is oscillatory.

-00
< liminf!
t--+=
1t 1
8
to
to
q(u)duds < lim sup !

t--+= t
1t 1
8
to
to
q(u)duds, (4.2.61)
320
Chapter 4
Theorem 4.2.9. In Theorem 4.2.8 condition (4.2.44) can be replaced by
lim sup -1()

t--+oo
.;
it
to
(is p(u)q(u)du )2 ds =
( )1 ( )
asp
to
00.
(4.2.62)
Let x(t) be a non oscillatory solution of equation (4.1.3), say,

~ to. Define wet) as in Theorem 4.2.8 and obtain (4.2.46).
Next, we consider the two cases: 1. l(to,oo) < 00 and 2. l(ta,oo) = 00.
The proof of Case 2 is exactly the same as of Theorem 4.2.8 - Case 2 and
hence omitted. Now, we consider
Proof.
x(t) > 0 for t
Case 1.
l~
p(s)q(s)ds = Kl
+ a(t)p(t)w'(t) -
ho
a(s)p(s)(w'(s2 f'(x(sds,
where Kl = -a(to)p(to)w'(to). Thus, if we define
K2
3K~ + 3
(L
OO
a(s)p(s)(w'(s))2 f'(X(S)dS) 2
then, we have
(1:
p(s)q(S)dS) 2 :s;
3K~ + 3(a(t)p(t)w'(t2
+ 3 (It: a(s)p(s)(w'(sWf'(x(sds) 2 :s; K2 + 3(a(t)p(t)w'(t2.

But, condition (4.2.42) implies that f'(x(tw(t)
1 for t ~ to. Hence, for t ~ to, we have
.;~) 1: a(s)~(s) (1: p(u)q(u)du

1
< K2 + ';(t)
< K2 + .;tt)
1:
lto
= f'(x(t Jx,,%) du/ feu) >
ds
a(s)p(s)(w'(s2ds
a(s)p(s)(w'(s2 f'(x(sw(s)ds
< K2 + .0(1) [max w(s)] {t a(s)p(s)(w'(s2 j'(x(sds,

." t
i.e.,
to~sst
lto
321
where K3 = 3 ft'; a(s)p(s)(w'(s))2 f'(x(s))ds > o. Now, as in Theorem

4.2.8, we obtain (4.2.48), and hellce we can choose aT:::: to so that
w(t) S; ~(t) for all
T, and
t::::
max w(s) S;
max w(s)
to:S,:ST
to:S'$!
for all
t:::: T.
t:::: T,
Thus, for
+ ~(t)
(4.2.63) gives
~tt) 1: a(s)lp(s) (1: p(u)q(U)dU)
ds
S;
[to~;~Tw(s) + ~(t)] .
K2 + ;:)
Therefore, we have
lim sup _(1)
t-+oo
~ t
it ()\) (1'
to
asp s
to
P(U)q(U)dU)2 ds
+ K3 <
K2
S;
Corollary 4.2.11.
by
In Corollary 4.2.9 condition (4.2.60) can be replaced
it
lim sup (: 1( )
t-+oo
00,
<,,1
_(1)
a s
to
(1' q(U)dU)
to
ds =
00.
11 l'
! it (f"
t
liminf t-+oo
and
lim sup
t-+oo
to
to
to
lto
q(u)duds > -
q(U)dU) ds
00
00,
(4.2.64)
Theorem 4.2.10.
by
In Theor!'lm 4.2.8 condition (4.2.43) can be replaced
liminf -1
t-+oo
tn
it
to
(t - stp(s)q(s)ds
>
-00,
(4.2.65)
where n> 1 is an integer and (a(t)p(t))' S; 0 for t:::: to.

to. Define the function w(t) as in Theorem 4.2.8 and obtain (4.2.46). Once again, we consider the two cases when a(t)p(t)(w'(t))2x
f'(x(t)) E 1 [to, 00) and a(t)p(t)( w'( t))2 f' (x(t)) if. 1 [to, 00). The proof
of the first case is exactly the same as in Theorem 4.2.8 - Case 1 and hence
omitted. Next, we consider the second case when J(to, 00) = 00. Define
Proof.
x(t) > 0 for
t::::
322
Cha.pter 4
c = infx>o f'(x) ixOO du/ feu). In view of condition (4.2.42), we have c> l.
Hence, we can choose an integer k;:: max{n,2} such that
k
- - < c.
(4.2.66)
k-1
As k;:: n, it is easy to verify that condition (4.2.65) implies that
rt (t - s)kp(s)q(s)ds
liminf 1k
t--too
.lto
> -
00.
(4.2.67)
Now, for all t;:: to from equation (4.2.46), we find
+ it(t -
i t (t - s)kp(s)q(s)ds
to
s)ka(s)p(s)(w'(s))2 f'(x(s))ds
to
i t (t - s)k(a(s)p(s)w'(s))'ds
to
- (t - to)ka(tO)p(to)u-'(to)
+ kit (t -
s)k-1a(s)p(s)w'(s)ds.
to
Thus, for t 2: to, we have

k1 i t (t - s)kp(s)q(s)ds
to
- (1- t:) ka(to)p(to)w'(to) + t~ [k 1: (t - s)k-1a(s)p(s)w'(s)ds
-1:
(t - s)ka(s)p(s)(w'(s))2 f'(X(S))dS] .
(4.2.68)
Next, in view of (4.2.66), we can choose a constant J.L such that
~
c
(_k_)
k-l
<
J.L
< l.
(4.2.69)
We claim that for every T;:: to, there exists at;:: T such that
t (t - sla(s)p(s)(w'(s))2 f'(x(s))ds > -kit (t - sl-la(s)p(s)w'(s)ds.

J.L
Otherwise, there is a T
to such that for t
t (t - s)ka(s)p(s)(w'(s))2 f'(x(s))ds
.lto
:s; '5..
T,
t (t - s)k-1a(s)p(s)w'(s)ds .
J.L .lto
(4.2.70)
323
By the Schwartz inequality and (4.2.70), and using the definitions of w

and c for t::::: T, we obtain
0::::
t (t -
ltD
s)k- 1a(s)p(s)w'(s)ds
:::: [1:(t-S)k a(S)p(S)(w'(S))2 f'(X(S))dS] 1/2
[1:(t-s)k-2;~~~~;~ dS] 1/2
k
]1/2 [1
]1/2
: : [ "ML
(t-s)k- 1a(s)p(s)w'(s)ds
~lo (t-s)k- 2a(s)p(s)w(s)ds
t
and hence, it follows that
t (t - s)k- 1a(s)p(s)w'(s)ds
ho
t::::: T,
But, for all
1.
::::
rt (t - s)k- 2a(s)p(s)w(s)ds. (4.2.71)
~~
we have
1,
(t - s)k- 2a(s)p(s)w(s)ds = -_k (t - to)k- 1a(to)p(to)w(to)

~
1
+ -_1k
1
1.t (t - s)k-1 [a(s)p(s)w'(s) + (a(s)p(s))'w(s)] ds.

to
Using the fact that (a(t)p(t))':::: 0 for t::::: to in the above inequality, we
get
rt (t _ s)k- 2a(s)p(s)w(s)ds
-1k (t - to)k- 1a(to)p(to)w(to)

-1
::::
+-k
_1
1
1.t (t - s)k- 1a(s)p(s)w'(s)ds

to
and so (4.2.71) for all t::::: T, gives
[ftC (k ~
1) _1] 1>-S)k-1 a(S)P(S)W'(S)dS :::: (t-to)k- 1a(to)p(to)w(to).
1) - 1] tk~l 1:
Therefore, by using (4.2.70), we find
[ftC (k ~
k (
:::: "M 1 - t;
But, J(to, 00)
= 00
1
lim k-1
t~oo
)k-1
(t - s)ka(s)p(s)(w'(s))2 f'(x(s))ds
(4.2.72)
a(to)p(to)w(to)
for
t::::: T.
implies
1.t(t - s)ka(s)p(s)(w'(s))2 f'(x(s))ds

to
00.
Chapter 4
324
On the other hand, (4.2.69) implies that ILC((k - l)/k) - 1 > O. Thus,
(4.2.72) leads to a contradiction. Hence, the claim follows and so we can
consider a sequence {tm}~=l with limm-too tm = 00 such that
tm
(tm - s)ka(s)p(s)(w'(s))2 j'(x(s))ds
to
>
tm
IL to
(tm - sl-la(s)p(s)w'(s)ds
(m
= 1,2", .).
Then from (4.2.68) it follows that

T1
tm
tm
(t m
+ (IL Now, since
s)kp(s)q(s)ds < -
t)k a(to)p(to)w'(to)
1 - -.2..
tm
tm
l)T
(tm - s)ka(s)p(s)(w'(s)f j'(x(s))ds,
1
tm to
m= 1,2,.
(4.2.73)
It
lim ~
(t - s)ka(s)p(s)(w'(s))2 j'(x(s))ds
t-too t to
= J(to, 00) =
00
and by (4.2.69) the constant IL < 1, the inequality (4.2.73) ensures that
=
This shows that

liminf k1
t-too t
It
to
(t - s)kp(s)q(s)ds
= -
- 00.
00,
Corollary 4.2.13. Assume that a'(t)::; 0 for t 2: to and condition

(4.2.42) is satisfied. If condition (4.2.60) holds and
liminf -1
t-too t n
It
to
(t - s)nq(s)ds > -
00
for some integer
n> 1, (4.2.74)

Corollary 4.2.14. Assume that condition (4.2.42) is satisfied. If condition
(4.2.74) holds and
lItiS
lim q(u)duds
t-too t to to
does not exist as a real number,
325

From Theorems 4.2.9 and 4.2.10 the following result is immediate.
Theorem 4.2.11. Suppose condition (4.2.42) is satisfied and there exists
a function p(t) E C 1([to,00),lR+) s11ch that condition (4.2.3) holds, and
(a(t)p(t))' -:: 0, a(t)p'(t) = p(t)p(t) for t 2': to. If conditions (4.2.62) and
(4.2.65) hold, then equation (4.l.3) is oscillatory.
Corollary 4.2.15. If conditions (4.2.42), (4.2.64) and (4.2.74) are satisfied, then equation (4.1.1) is oscillatory.
Next, we present a more general result.

Theorem 4.2.12.
Suppose condition (4.2.42) is satisfied and assume
that there exists a function p(t) E C 1([to, 00), JR+) such that a(t)p'(t) =
p(t)p(t), (a(t)p(t))' -:: 0 for t 2': to and condition (4.2.3) holds. If
condition (4.2.65) holds, and there exists an integer n > 1 such that
lim sup
t-->oo
~n rt (t - s)n-2a(s)p(s)~(s)ds
t lto
and
1
lim sup --;-
t-->=
it
(t - s
to
<
p( s )q( s )ds
00
(4.2.75)
00,
(4.2.76)

Let x(t) be a nonoscillatory solution of equation (4.l.3), say,
x(t) > 0 for t 2': to. We define w(t) as in Theorem 4.2.8 and obtain
(4.2.46). Next, we consider the two cases: 1.
J(to,oo) < 00 and 2.
J(to, 00) = 00.
Proof.
The proof of Case 2 is exactly the same as in Theorem 4.2.10 - Case 2.

Thus, we consider Case 1. Proceeding as in Theorem 4.2.8 - Case 1, there
exist a constant C1 > 0 and a T 2': to such that
w(t) -::
C1~(t)
for
t 2': T.
(4.2.77)
Define y(t) = a(t)p(t)w'(t), t 2': T so that equation (4.2.46) can be written

as
y'(t) = p(t)q(t)
+ a(t)lp(t) y2(t)j'(X(t)),
t 2': T.
(4.2.78)
Using (4.2.42) and (4.2.77) in equation (4.2.78), we get
y'(t) 2': p(t)q(t)
+ c1a(t)~(t)~(t) y2(t)
for
t 2': T.
(4.2.79)
Chapter 4
326
Multiplying (4.2.79) by (t - s)n and integrating by parts, we find
it
T
(t - s)np(s)q(s)ds
- n
it
or
t [(
(t-s)n
C1a(s)p(s)~(s)
+ -C11
it
T
(t - s)n-1y(s)ds
)1/2 yes) -
f\t-s)np(s)q(s)ds - n 2C1
iT
(t - st ( ) ( ) ( ) y2(s)ds
asp s ~ s
:s: - (t -
T)ny(T),
n(t_s)(n-2)/2]2
2(C1a(s)p(s)~(s-1/2
ds
t(t-s)n-2a(s)p(s)~(s)ds :s: -(t-T)ny(T).
iT
tn
t(t-s)np(s)q(s)ds < n2cl
iT
4t n
t(t-s)n-2a(s)p(s)~(s)ds- (l_~)n yeT).
iT
(4.2.80)
Taking limsup on both sides of (4.2.80) as t -+ 00 and using (4.2.75), we
obtain a contradiction to condition (4.2.76). This completes the proof.
Corollary 4.2.16.
(4.2.74) holds and
lim sup -1
t .... oo t n
it
to
Suppose condltion (4.2.42) is satisfied. If condition
(t - stq(s)ds =
00
for some integer
n > 1,
(4.2.81)

Corollary 4.2.17. If conditions (4.2.74) and (4.2.81) are satisfied, then
Remark 4.2.3.
1. We note that (4.2.16) implies the first part of (4.2.61) and inductively
condition (4.2.74).
2. The condition limt-too
(4.2.17) implies (4.2.61).
It: q(s)ds =
00
implies (4.2.17), and condition
3. As an application of the Schwartz inequality condition (4.2.17) implies

(4.2.64). We also note that condition (4.2.30) implies (4.2.17) which is
equivalent to limt-too(l/t) ftto(t - s)q(s)ds = 00.
4. Extensions of the above theorems which involve the function (t - s)n
can be obtained by replacing it with the more general function H(t, s)
satisfying the same conditions listed earlier (see Chapters 2 and 3).
327
5. It would be interesting to obtain oscillation criteria for equation (4.1.3)

similar to those of Theorems 4.2.8 ~ 4.2.12 without the assumption that
(J(t) = 0, i.e., letting pet) = exp (t(p(s)/a(s))ds).
In the following results, we remove the assumption on
condition (4.1.28).
l' (x),
i.e., the
Theorem 4.2.13. Suppose conditions (F2) and (4.1.9) hold and Q(t) =
oo q(s)ds exists, and liminfHoo
Q(s)/a(s)ds > -00, and either
(Q+(s))2/(a(s)6(s))ds = 00, or for every constant c> 0,
It
It";
It:
lim sup
t ..... oo
where 6(t)
00
to
ass
Itto ds/a(s),
(~(U)(Q+(U))2 du jdS
_(1) [Q(s)+cjoo
al 1 ~l
00,

x(t) > 0 for t 2 to. It follows from Lemma 4.1.3 that (4.1.50) holds.
Proceeding as in Theorem 4.2.2 ~ Case 1 (take pet) = 1 and (J(t) = 0),
there exist a T 2 to and a constant Cl > 0 such that
Proof.
1'(x(t)) 2
~;tt)
for
00
Cl
t2
(4.2.82)
T.
a(s) ( x'es)
6(s) f(x(s))
)2 ds for t 2 T.
Remark 4.2.4. We note that Theorem 4.1.8 includes Theorem 4.2.13.

Now as in Theorem 4.1.9, we define
ho(t)
00
y'a(t)6(t)
1
= 1
00
q(s)ds,
6(t)
to
ds
a(s)'
([ho(s)]+)2 ds
00
([hO(S)
y!a(s)~l(S)
hn(S)]
+)
ds
for n
= 1,2,
where c is a positive constant.

Theorem 4.2.14. Suppose conditions (F 2 ), (4.1.9) and (4.1.10) are satisfied and there exists an integer N such that hn(t) exists for n =
328
Chapter 4
0,1, ... ,N. If for ever constant c > 0, and for all sufficiently large t,
ho(t) + chN(t)/ a(t)6(t);::: 0 and
/ 00
y'a(s)6(s)
[ho(S)
y'a(s)6(s)
hN(S)] ds
=,

x(t) > 0 for t;::: to. As in Theorem 4.2.13, we obtain (4.2.82) for t;::: T.
The rest of the proof is exactly the same as that of Theorem 4.1.9.
Proof.
4.3. Existence of N onoscillatory Solutions

and Comparison Results
In this section first we shall present sufficient conditions which guarantee the existence of a nontrivial nonoscillatory solution of (4.1.1), then
we shall provide comparison theorems similar to those of Hille-Wintner for
second order equations of the type (4.1.2). This will be followed by several
oscillatory theorems involving comparison with related linear equations.
This will enable us to employ the results of Chapter 2 to obtain interesting
oscillation criteria for nonlinear equations.
4.3.1. A Necessary Condition for Oscillation

We shall prove the following result.
Theorem 4.3.1. Let f(x) and q(t) be such that
there is a nontrivial compact interval I in which xf(x) > 0 for
x -=1= 0 and f'(x) exists at each point, is nonnegative and bounded,
(4.3.1)
q(
t)
is locally integrable and Q( t) = OO q( s )ds exists (possibly
{
infinite) with liminft---+oo OO Q(s)ds > - = for all t
(4.3.2)
and
{
It
It
/00 (1 00 (Q-(U2 dU)
where Q-(t)
= Q+(t) - Q(t).
ds <
=,
(4.3.3)
Then,
/00 [Q(S) + 1= Q2(U)dU] ds < =
(4.3.4)
implies that there exists a nontrivial nonoscillatory solution of equation

(4.1.1).
329
Proof. Let the hypotheses of the theorem hold, so that
(Q is absolutely integrable in view of (4.3.2)).
To prove the existence of a nontrivial nonoscillatory solution of (4.l.1)

we shall employ the Schauder fixed point theorem. Without loss of generality, let I = [b, c] with 0 E such that [0' - E, 0' + E] c [b, c]. We define the constants
A, AI, B and the function P(t) as follows
A
sup f(u),
bC:;uC:;c
Al
sup f'(u)
(4.3.5)
bC:;uC:;c
(4.3.6)
Now, we choose T > 0 so large that
(4.3.7)
for all s, t:2: T,
and
lXo P(s)ds
:::;
(4.3.8)
E.
Define X to be the space of bounded, absolutely continuous functions on

[T, 00) whose derivatives are almost everywhere bounded in absolute value
by some multiple of P(t).
Let 1111= denote the uniform norm. We make X into a Banach space
by introducing the norm
For n = 1,2, . .. define Xn to be the subset of functions x in X such
that Ilx - 0'1100 :::; E, IIp-Ix'lloo:::; 1, and
Ilx'(t 2 )
x'(tl)11 :::; A
11:
q(s)dSI
+ it 2
tIl
for all t l , t2 E [T, T + n) and x is constant for t:2: T + n. Then, Xn

is a convex subset of X, and by the Ascoli theorem Xn is compact.
For x E Xn define the map Tn by
(Tnx)(t) =
0'
-1 1
0' -
00
DO
00
q(u)f(x(u))duds
JT+n s
for
T:::;
DO
q(u)f(x(u))duds
for
t:::;
+n
+ n:::; t.
330
Chapter 4
We note that since x(t) E Xn is constant on [T+n, (0), the integrability

of q and IQI imply that Tn is a well-defined map of Xn at least into the
space of absolutely continuous function with almost everywhere bounded
derivatives.
We shall now show that Tn is a continuous map of Xn into itself. Let

X n . Then for T -s; t < T+n, we have
(Tnx)'(t) =
00
q(s)f(x(s))ds = f(x(t))Q(t)+
00
Thus, by (4.3.5) - (4.3.7) it follows that
I(Tnx)'(t)1
< AIQ(t)1 +
< AIQ(t)1 +
~B
(1
00
00
f'(x(s))Q(s)x'(s)ds.
1
(1
00
AlIQ(s)1 [AIQ(s)1 +B
00
Q2(s)ds + AlB
00
Q2(S)dS)
-s; P(t).
IQ(s)ldS)
Therefore, Tn(Xn) C X. Furthermore, if T -s; tl < t2 < T
I(Tnx)'(t2) - (Tnx)'(tl)1
jI(x(t d )
11:
< A
1:
11:
1:
+ n,
we have
q(s)f(x(s))dSI
q(s)ds + lt2 J'(x(s))x'(s)
q(s)dsl
Q2(U)dU] ds
AlP(s)
11
00
(l t2 q(U)dU) dsl
q(U)dul ds.
Using (4.3.7) in the above inequality, we obtain
For T -s; t -s; T
ITnx(t) - 01
-s;
+ n,
n
+ [AIQ(s)1
we find
11 1
00
+B
00
q(U)f(x(u))dUdSI
1 Q2(U)d~
00
ds
+ l:/(x[T+n])
11
00
q(u)dul ds.
Thus, by (4.3.8), we get ITnx(t) - 01 -s; ftOO P(s)ds -s; E. Since Tnx(t) is
constant for t 2 T + n, this shows that Tn(Xn) C X n .
331
Ilx - yll
Now, let X,y E Xn with
ao(t5)
=
b
 0 is a constant). Define
IJ(u)-J(v)1 and al(8) =

b
<
sup
u, u
<
a.
+1':
1f'(u)-f'(v)l
iTu - vil:'O .s
Then, we have
I(Tnx)' - (Tny)'I(t)
00
11 [J(x(s)) - J(y(s))]q(s)dsl
IJ(x(t)) - J(y(t))IIQ(t)1
+ 11[J'(x(s))x'(s)- J'(y(s))Y'(s)]Q(s)dsl
IJ(x(t)) - J(y(t))IIQ(t)i
ao(t5)IQ(t)1
ao(t5)IQ(t)1
+ (B
00
IJ'(x(s)) - f'(y(s))lx'(s)IIQ(s)lds
[:xJ 1f'(y(s))llx'(s) _ y'(s)IIQ(s)lds

+
00
+A 1 t5 ( AIQ(s)1
00
[a 1 (8) ( AIQ(s)1
+B
00
(1=
00
Q2(U)dU)
Q2(U)dU)] IQ(s)lds
+ (al (8) + 8A 1 )
Q2(S)dS)
+B
[A
00
Q2(s)ds
IQ(s)ldS) ]
ao(t5)IQ(t)1 + (al(8) + 8Al)(A + EB)
00
Q2(s)ds
b(8)Q(t),
say, where b( 8) -+ 0 as 8 -+ O. Integration of the last inequality gives
ITnx - TnYI
b(t5)
00
00
I(Tn x)' - (Tny)'I(s)ds
P(s)ds
+ b1 (8)
00
P(s)ds
+ ITnx -
TnYI(T + n)
+ b1 (8)),
E(b(8)
say,
where b1 (8) has the same property as b(8), and so IITnx - TnYl1 ~
(b(t5) + b1 (8))(1 + E). Hence, it follows that Tn is continuous. Now, we
can apply the Schauder fixed point theorem to obtain a function Xn E Xn
such that Tnxn = x n . Thus, for T ~ t < T+n, x~(t)+q(t)J(xn(t)) = O.
The sequence {x n } is equicontinuous on [T, 00 ). It follows by a standard
diagonal argument that there is a subsequence {x nk } which converges uniformly on compact intervals to a function x which is nontrivial, bounded
nonoscillatory solution on [T, (0) of equation (4.1.1).
Chapter 4
332
We can combine Corollary 4.1.4 and Theorem 4.3.1 to obtain the following necessary and sufficient condition for the oscillation of (4.1.1).
It
Corollary 4.3.1. Let Q(t) = OO q(s)ds exist with Q(t) 20 for t 2 to,
and let f be strongly nonlinear, i.e., f satisfies the conditions
C 1(lR,lR), xf(x)
> 0 for xi- 0, f'(x) > 0 for xi- 0, and
(f1)
(f2)
liminfixi-4oo f'(x) > 0 and condition (4.1.5) holds.
Then equation (4.1.1) is oscillatory if and only if
If we remove the sign condition on Q,

(4.1.1) to be oscillatory is
then a necessary condition for
4.3.2. Comparison of Nonlinear Equations

of the Same Form
The classical Sturm comparison theorem for the second order linear
differential equations
(a1(t)x'(t))'
+ q1(t)X(t)
(a(t)x'(t))' + q(t)x(t)
0,
(4.3.9)
(4.3.10)
is stated as follows: If a, aI, q, q1 are continuous functions with 0 <

a(t) ~ al(t), q1(t) ~ q(t) for all t in [to, (0), and if equation (4.3.9) is
oscillatory, then (4.3.10) is also oscillatory.
A clever application of the Sturm theorem gives the so~called Hille~
Wintner comparison theorem which is in terms of the integrals of the coefficient functions qI, q.
Theorem 4.3.2. Let a(t) = al(t)
= 1,
and let
Q(t)
Q1(t)
00
ql(s)ds,
00
q(s)ds
(4.3.11)
exist with 0 ~ Ql(t) ~ Q(t) for all t in [to, (0). If equation (4.3.9) is
oscillatory, then (4.3.10) is also oscillatory.
This result has been extended to the following form:
333
Theorem 4.3.3. Let a(t) be bounded above on [to, 00). Let Q1 (t), Q(t)
in (4.3.11) exist and let 0 < a(t) ::; a1(t), 0::; IQ1(t)1 ::; Q(t) for all t in
[to, 00). If equation (4.3.9) is oscillatory, then (4.3.10) is also oscillatory.
We shall extend Theorems 4.3.2 and 4.3.3 to nonlinear equations of the

form
(4.3.12)
(a1(t)x'(t))' + q1(t)f(x(t)) = 0,
(a(t)x'(t))'
+ q(t)f(x(t)) = o.
(4.3.13)
Theorem 4.3.4. Let a, aI, q, q1 be continuous 011 [to, 00) such that
Q(t) and Q1(t) defined in (4.3.11) exist and such that 0 < a(t) ::; a1(t),
IQ1(t)1 ::; Q(t) for all t E [to, 00). Assume that the function f satisfies
the following conditions:
f(x) is continuously differentiable, xf(x) > 0 for all x # 0,

f'(x) > 0 for x # 0, and either
(ii) f'(x) is nondecreasing on [0,00) and is nonincreasing on (-00,0],
(i)
or
(iii) liminf 1xl -4oo f'(x) > 0 and condition (4.1.5) holds.
If equation (4.3.12) is oscillatory, then (4.3.13) is also oscillatory.
Proof. It will be convenient to separate the proof into the following three
cases:
(I)
(II)
If I
Joo du/a(u) = 00,
Joo du/a(u) = 00,
convex (conditions (i) and (ii)),
f strongly superlinear (conditions (i) and (iii)),
(III) Joo du/a(u) < 00.

Case (I). Suppose (4.3.13) is not oscillatory. Then there is a solution x(t)
of (4.3.13) which is eventually of one sign, and without loss of generality,
we assume that x(t) > 0 on [to, 00). Let w(t) = a(t)x'(t)/f(x(t)) to
obtain the Riccati equation w'(t) = -q(t) - w 2 (t)f'(x(t))/a(t). Then for
to ::; t ::; T, we find
w(t)
w(T)
iT iT
t
q(s)ds +
a(s) w 2 (s)f'(x(s))ds.
(4.3.14)
Letting T -+ 00 in (4.3.14), the second term on the right-hand side has the
nonnegative limit Q(t), whereas the third term has either a nonnegative
limit, or tends to 00. Hence, it follows that limT-4oo w(T) = (3, where
- 00 ::; (3 < 00.
We shall show that 0::; (3 < 00. Suppose on the contrary that - 00 ::;
(3 < O. Then for sufficiently large t, say, t?: l, we have a(t)x'(t) < O.
If for some E, a(t)x'(t)::; -E < 0 for t?: t* ?: l, then it follows
that x(t) - x(t*) ::; -E
ds/a(s) -+ -00 as t -+ 00, which contradicts
Itt.
Chapter 4
334
x(t) > 0 on [to, 00). Thus, we must have limsuPt-+oo a(t)x'(t) = O.

We obtain a contradiction as follows: Choose tn > f with tn --+ 00
as n --+ 00, so that for sufficiently large n and for f::::; t ::::; tn,
-lin = a(tn)x'(t n ) > a(t)x'(t). Integrating equation (4.3.13) from t to
tn, f::::; t < tn for n large, we obtain
0= a(tn)x'(tn) - a(t)x'(t)
=
(l
tnq (S)ds) f(x(t
= V(t) -
tn
tn
q(s)f(x(sds >
(1\(U)dU )f(X(S
,, q(s)f(x(sds
(x'(;~~'(~~~s
)dS
,, H(s)V(s)ds,
where
V(t) =
(l tn q(S)dS) f(x(t
and
H(t) = -x'(t)f'(x(t)) > O.

f(x(t
Let W(t) = Ittn H(s)V(s)ds. Then the inequality V(t) < Ittn H(s)V(s)ds
gives W'(t) = -H(t)V(t) > -H(t)W(t) from which it follows that
and hence 0 = W(tn) exp (Itt,. H(s)ds) > W(t) exp (I; H(s)ds) , which
implies that W(t) and thus V(t) is negative on [f, tn). But this in
turn implies that Ittn q( s )ds < 0 for f::::; t < tn for n sufficiently
large, contradicting that 0::::; Q(t) = Itoo q(s)ds. Hence, on taking limit in
(4.3.14), we find
w(t)
where 0::::; (3 <
have
00
(3 + Q(t)
00
a(s) w 2 (s)f'(x(sds,
as claimed. Now, since x'(t)1 f(x(t
(t) ~ _
f(u) -
lto
(4.3.15)
= w(t)la(t),
we
(w(s) ds
a(s)
,
where 0: = x(to) > O. Denoting I:(t) dulf(u) by n(x) for x> 0 and
the inverse function of n by r, we get
Oscillation theory for superlinear diHerential equations
335
and hence (4.3.15) can be rewritten as
Conditions (i) and (ii) imply that r is monotone increasing from some
interval (-oo,c) to lR+ (c maybe +00) and that we can write (4.3.16)
as
w = M(w),
(4.3.17)
maps the set C = {u E C([to,oo),lR+) with O:'S: u(t) :'S:
[to,oo)} into itself. lVI is monotone in the sense that lVIu < lVIv
for any u, v E C with u < v under the na.tural ordering of C. Now
where M
w(t), t
consider the integral equation
which we may write as
z = L(z).
(4.3.19)
Since IQI(t)l:'S: Q(t) and 0 < a(t):'S: al(t) on [to, 00), we find that L
is a self map on B = {u E C([to, 00), lR+) with QI(t):'S: u(t) :'S: z(t), t E
[to,oo)}. Clearly, B is a closed convex subset of C = C([to, 00), lR)
the Fnchet space of continuous real-valued functions on [to, 00) with the
compact-open topology. Let z E B. Then, we have
Iz(t)1 :'S: w(t),

and for to:'S: s
< t :'S: T,
I(Lz)(t) -(Lz)(s)1 :'S:
t E [to, 00)
ILz(t)l:'S: w(t),
ilt
ql(u)dui +
ilt
(4.3.20)
q(u)dui + Iw(t)-w(s)l (4.3.21)
From (4.3.20) and (4.3.21) it follows that the functions in the image L(B)
of B are uniformly bounded a.nd equicontinuous on compact subintervals
of [to, 00). We claim that L is continuous on B. For this, for z E C
define <1>( z; s) by
<1>(z;s) =
(Z2(S))
f' (r [1 (~) dU])
al(S)
to
al(U)
8
so that
(Lz)(t)
Let z, Zn E C be such that

subintervals of [to, 00). Let
TI ~ tl so that
(J+QI(t)+
Zn ---}
00
<1>(z;s)ds.
(4.3.22)
z as n ---} 00 uniformly on compact

E > 0
be given and choose
= [to, tIl,
iT}roo (wa(s)(S)) f' (r [1to

2
(w(u)) dU]) ds < -3E

a(u)
(4.3.23)
Chapter 4
336
Noting that 1>(zn; s) -+ 1>(z; s) as n -+ 00 uniformly on [to, TIl, using

(4.3.21), (4.3.22), (4.3.23) and the monotonicity of f' 0 r, we obtain for
tEl,
~ 11Tl (1)(z; s) -1>(zn; S))dSI
I(Lz)(t) - (Lzn)(t) 1
+ {= (11)(z; s)1 + 11>(zn; s)l)ds <
-3E
iTl
+~ <
3
E,
for n sufficiently large. Thus, LZn -+ Lz as n -+ 00 uniformly on

compact subintervals of [to, (0). It follows that L is continuous on B
and so we can apply Schauder-Tychonov's fixed point theorem to deduce
the existence of a fixed point Z of (4.3.20). Then, we find
is a nonoscillatory solution of equation (4.3.12) on [to, (0). This completes

the proof for case (I).
Case (II). A change of variables s =
forms equation (4.3.13) into
d2 y
ds 2
It: dr/a(r), y(s)
+ a(t(s))q(t(s))f(y(s)) =
= x(t(s)) trans-
(4.3.24)
O.
The oscillation properties of the equation are invariant under this transformation. Let r(s) = a(t(s))q(t(s)), and use the substitution u = t(r), to
obtain
1=
{=
l= (1=
R(s)
r(r)dr
iT
1=
(=
a(t(r))q(t(r))dr
R(s)ds =
iT
R 2 (r)dr) ds
Q(t(s))ds =
1=
1=
t(T)
q(u)du
t(s)
Q(t(s)) > 0,
(4.3.25)
(Q((u))) du,
au
1(:) a(~) ([= (~~~~))
(4.3.26)
d7]) duo (4.3.27)
It follows from Corollary 4.3.1 and (4.3.24) - (4.3.27) that equation (4.3.13)
is oscillatory if and only if
J=
1 [ Q(u)
a(u)
1= (Q
2
(7])) d7] ] du =
a(7])
00.
(4.3.28)
Similarly, we find that a necessary condition for (4.3.12) to be oscillatory

is that
al(u) IQI(U)I +
1 (Qi(7]))
00
al(7])
d7] du =
00.
(4.3.29)
Oscillation theory for superlinear differential equ8,tions
:s:
o:s:
337
:s:
Since 0 < a(t)

a1(t) and
IQ1(t)1 Q(t), from (4.3.28) and (4.3.29)
it follows that the theorem holds for case (II).
Case (III). We claim that condition (i) and the existence of the integral
Q1(t) imply that (4.3.12) has a nonoscillatory solution. For this, we use
the transformation s = g(t) = (ftOO dT/adT)) -1, y(s) = sx(t). Now, if
we denote g( to) by So and the inverse function of 9 by h, then this
transformation changes (4.3.12) into
d2 y2
ds
where rds)
+ r1 (s )f (y(s))
-_
s
0,
SE[to,::x:J)
(4.3.30)
= a1(h(s))Qdh(s))/s3, Choose a constant K 2: 1 such that
o:s:
f'(x):S: K
whenever
0:S:x:S:2
(4.3.31 )
and choose To 2: max{l, so}, so that
(4.3.32)
Consider the integral equation
(4.3.33)
which we write as y
= F(y). Let
Thus, R1 (s) is well defined and satisfies
(4.3,34)
Next, we denote [To, 00) by J and let X to be the Banach space
of bounded continuously differentiable functions y on J such that ty'
is bounded on J In the space X we introduce the norm as Ilyll =
IIYlloo + Ilty'lloo, where I 1100 denotes the supremum on J
Let 13 be the closed ball in X with center 1 and radius 1/2. We shall
show that F maps 13 continuously into a compact subset of 13, so that
the Schauder fixed point theorem can be used to guarantee the existence
of y* E 13 such that F(y*) = y*. The function y*(s) will then be a
Cbapter 4
338
nonoscillatory solution of (4.3.30), which in turn will give a nonoscillatory

solution of equation (4.3.12). Now, in what follows we let s ~ To ~ 1. If
y E B, then we have Iy(s) -11::::; 1/2 and Isy'(S)I::::; 1/2, and therefore
Noting that 0::::; y(s)/s < 2/s < 2 and using (4.3.31)
and so
I[F(y)](s)
Q*
-11 ::::; 6K -;- ::::;
(4.3.34), we get
(4.3.37)
From (4.3.36) and (4.3.37), we find that F(y) E B and hence F maps
y, z E B with lIy - zll : : ; E. We have
B into itself. Now, let
<
11
[I (Y~)) - 1(z~)) ] dTI

IRl(S)1 JI (y~)) - 1 (z~S)) 1+ 1 IR1(T)IIU(T)ldT,
I[F(y)]'(s) - [F(z)]'(s) I =
00
rl
(T)
00
(4.3.38)
where
U(T)
f'
(Y~)) (Y~))' - f' (z~)) (z~))'
Using (4.3.31) and the mean value theorem, we obtain
< KIY(s) - z(s)1 <

s
Kf
S
(4.3.39)
Writing U(T) as
[t (Y~)) - f' (z~))] (y~))' + f' (z~)) [yl(T)~ZI(T) + Z(T);;y(T)] ,

we can use (4.3.31) and (4.3.35), to get
IU(T)I ::::;
(:2)'(~) +2K;2
<
:2C!(f) + Kf),
(4.3.40)

where , is the modulus of continuity of
(4.3.39) and (4.3.40) in (4.3.38), we find
I[F(y)]'(s) - [F(z)]/(s)1
~
<
f'
339
on [0,2]. Using (4.3.34),
22KQiE + 4Qi(r(E)
+ KE)
(1 d:)
00
S s T
~[Qi(2KE
+ ,(E))]
S
(4.3.41 )
and therefore
I[F(y)](s) - [F(z)](s)1
-[Qi(2KE + ,(E))].
(4.3.42)
It follows from (4.3.41) and (4.3.42) that IIF(y)-F(z)11 ~ 4Qi(2KE+,(E))

and hence F is a continuous mapping of 13 into itself. Finally, let y E 13
and To ~ T ~ T. Then, we have
I[F(y)]'(s) - [F(Y)]/(T)I =
lis
Tl(/L)f
(Y~)) dTI ~
2K
is
!rI(/L)ld/L.
(4.3.43)
From (4.3.43), it follows that the image F(13) of 13 under F consists
uniformly bounded functions on J, with uniformly bounded derivatives on
J that are equicontinuous on compact subsets of 1. Thus, we can apply
the Ascoli-Arzela theorem to deduce that F(13) has a compact closure
in X. Now, the Schamler fixed-point theorem completes the proof of case
(III).
Remark 4.3.1. In equations (4.3.12) and (4.3.13) the function f(x) = x

satisfies conditions (i) and (ii), whereas f(x) = xO: where Q > 1 is
the ratio of odd integers satisfies conditions (i), (ii) and (iii). Hence, the
following corollary is immediate.
Corollary 4.3.2. Theorem 4.3.2 for (4.3.9) and (4.3.10) holds without
the restriction that a( t) is bounded, and it extends to nonlinear equations
of the form (4.3.12) and (4.3.13) where f(x):= xO: and Q > 1 is the ratio
of odd integers.
4.3.3. Linearization of Oscillation Theorems

In Chapter 2, we have shown via certain Riccati transformations that
the oscillatory behavior of equations of type (4.l.3) is equivalent to the
oscillatory behavior of linear differential equations provided conditions (i)
- (iii) on the coefficients of (4.l.3) and the condition (4.l.28) are satisfied. Here, via comparison we shall relate the oscillation of the superlinear
equations to that of certain linear differential equations.
Theorem 4.3.5. Suppose condition (4.1.28) holds. If the equation
(a(t)y/(t))/
+ kq(t)y(t)
(4.3.44)
Chapter 4
340
Let x(t) be a nonoscillatory solution of (4.1.2), then wet) =

-a(t)x'(t)/ f(x(t)) satisfies the Riccati differential equation w'(t) = q(t) +
w2(t)f'(x(t))/a(t) for some to;::: O. Let pet) = a(t)/f'(x(t)). Then the
Proof.
Riccati equation
w'(t) = q(t)
+ pet) w 2(t)
has a solution on [to, 00). It is well-known that this is equivalent to the

nonoscillation of the linear equation
(p(t)y'(t))'
+ q(t)y(t) =
0,
t;::: O.
(4.3.45)
By assumption pet) :s; a(t)/k, and hence in view of the Picone-Sturm

theorem the equation a(t)/k)x'(t))' + q(t)x(t) = 0, is nonoscillatory,
which contradicts the hypothesis.
We are now in the position to prove the following results.

Theorem 4.3.6. Suppose condition (F 2 ) is satisfied and let the functions
pet), f3(t) and ~(t) be as in Theorem 4.2.1, and conditions (4.2.3) - (4.2.5)
hold. If for every constant k > 0 the linear equation
(vtt) y'(t)), + kB(t)y(t) = 0
(4.3.46)
B(t)
and vet)
p(t)q(t)
+k
f3(t)
)'
k ( f3(t) )2]
[( 2a(t)p(t)v(t)
- 4v(t) a(t)p(t)
l/(a(t)p(t)~(t)), t ;:::
to then equation (4.1.3) is oscillatory.
Proof. Let x (t) be a nonoscillatory solution of equation (4.1.3), say,

x(t) > 0 for t;::: T ;::: to. Define wet) = p(t)a(t)x'(t)/ f(x(t)), t;::: T and
proceed as in Theorem 4.2.1 to obtain ~.2.7) and conclude that (4.2.8) and
(4.2.9) are satisfied with K = f3(T) ix(T) du/f(u). Now, we consider the
following two cases:
Case 1. l(T,oo) < 00. As in Theorem 4.2.2, we obtain (4.2.25) which takes
the form
w'(t) :s; - p(t)q(t)
+ a~)(;~t) wet) -
ev(t)w 2 (t),
t;::: T.
= wet) - f3(t)/(2ea(t)p(t)v(t)), t ;::: T. Then, we find that z'(t):s;

-B(t) - ev(t)z2(t) with k = l/e for t;::: T. Now, in view of Lemma 2.2.1
Let z(t)
it follows that equation (4.3.46) is nonoscillatory, which is a contradiction.
341
Case 2. I(T,oo) = 00. Exactly as in Theorem 4.2.1, we arrive at the

contradiction limt ..... oo x(t) = -00. This completes the proof.
Theorem 4.3.7. If conditions (4.2.75) and (4.2.76) in Theorem 4.3.12 are

replaced by that the equation
(4.3.47)
(vtt)Y'(t)), +kp(t)q(t)y(t) = 0
is oscillatory for every constant k > 0,
where v(t)
l/(a(t)p(t)~(t))

x(t) > 0 for t?: T ?: to. Proceeding as in Theorem 4.2.12, we obtain
Proof.
(4.2.77) and the equation (4.2.78). Using condition (4.2.42), we find
f'(x(t)) ?:
~~t)
for
(4.3.48)
t?: Tl ?: to,
where c is a positive constant. As in Theorem 4.3.5 since y( t) satisfies

(4.2.78) the corresponding linear equation
)'
( a(t)p(t),
f'(x(t)) Y (t) + p(t)q(t)y(t) = 0
(4.3.49)
is nonoscillatory. Now, by (4.3.48) and as an application of the PiconeSturm comparison theorem to (4.3.49), we conclude that the equation
(y'(t)/v(t))' + cp(t)q(t)y(t) = 0 is nonoscillatory, which contradicts the
given hypothesis.
Remark 4.3.2.
1. We can proceed further to obtain more comparison results similar to

those of Theorems 4.3.5 - 4.3.7 with different sets of conditions on the
coefficients involved in the equations considered. The details are left to the
reader.
2. We can use the results of Chapter 2 on linear equations to obtain interesting oscillation criteria for superlinear equations. These details are also
left to the reader.
4.4. Oscillation Criteria for Certain

Here, we shall consider the nonlinear differential equation
(a(t)lx'(t)I"-lX'(t))'
+ q(t)f(x(t))
0,
(4.4.1)
Cha.pter 4
342
where
(i) O! is a positive constant,
(ii) aCt) E C([to,oo),IR+) and q(t) E C([to, 00), IR),
(iii) f E C(IR, IR), f is non-decreasing and xf(x) > 0 for x
-I- O.
The results we shall prove extend Lemmas 4.1.2 and 4.1.3 and Theorems
4.1.7 - 4.1.9. For this, we need the following lemma.
Lemma 4.4.1. Let K(t, s, x) : IR x IR X IR+ -* IR be such that for all
fixed t, s the function K(t, s,') is nondecreasing. Further, let pet) be
a given function and u(t), vet) satisfy for t::::: to,
u(t) :::::
(~) pet) +
and
vet) = pet)
it
K(t, s, u(s))ds
(4.4.2)
to
it
K(t,s,v(s))ds.
(4.4.3)
to
If v*(t) is the minimal (maximal) solution of the inequality (4.4.2), then

u(t) ::::: (~)v*(t) for all t::::: to.
Lemma 4.4.2. Let x(t) be a positive (negative) solution of equation

(4.4.1) for t E [to,b], b> to. Suppose there exist tl E [to,b] and (3 > 0
such that for all t E [tl' b],
_ a(to)!x'(to)i",,-lx'(t) +
f(x(to))
it
q(s)ds +
to
it'
to
a(s) !X~(S)!O+l J'(x(s))ds ::::: (3.

f (x(s))
(4.4.4)
Then the following holds
a(t)!x'(t)!o-lX'(t) ~ (:::::) - (3f(X(tl))
for
[iI, b].
(4.4.5)
Proof. Equation (4.4.1) is equivalent to
w'(t)
f(x(t))
q(t),
(4.4.6)
where wet) = a(t)!x'(t)!o-lx'(t). In view of (4.4.6) it is clear that
( wet) )'
f(x(t))
w'(t)
f(x(t))
w(t)J'(x(t))x'(t)
J2(x(t))
!x'(t)!O+l J'(x(t))
-q(t) - aCt)
J2(x(t))
.
---
(4.4.7)
343
Let t E [tl' b]. Then on integrating (4.4.7) from to to t and then applying
(4.4.4), we find
rt
w(to)
- f(x(t o))
w(t)
f(x(t) )
---
(3 +
2:
1
t
tl
rt
+ ito q(s)ds + ito a(s)
a(s)
Ix'(s)ICt+1 f'(x(s))
J2(x(s))
ds
IX'(s)i<>+1 f'(x(s))
J2(x(s))
ds >
o.
(4.4.8)
Case 1. Suppose x(t) is positive. Then it follows from (4.4.8) that --w(t) >
0, or x'(t) <
for t E [tl, tl. Let u(t) = -w(t) = a(t)lx'(t)Ii3. Then,
(4.4.8) provides
2:
u(t)
(3f(x(t))
1t
+ 1t
+
tl
(3f(x(t))
f(x(t)) Ix'(s)If'(x(s)) u(s)ds

J2(x(s))
(4.4.9)
K(t, s, u(s))ds,
tl
where the function K
K(t,s,y)
is defined by
Ix'(s)If'(x(s))
J2(x(s)) y, t,s E [tl,b] and y E JR+. (4.4.10)
f(x(t))
It is obvious that for fixed t, s, K(t, s,) is nondecreasing. With p(t) =

(3f(x(t)) an application of Lemma 4.4.1 to (4.4.9) yields
u(t) 2: v*(t)
for
t E [tl, b],
(4.4.11)
where v*(t) is the minimal solution of the equation
v(t)
(3f(x(t))
(3f(x(t))
+
+
K(t, s, v(s))ds
itl
t f(x(t)) Ix'(s;If'(x(s))
ds
it
f (x(s))
(4.4.12)
provided v(t) E JR+ for all t E [tl, b].

( ~)'
f(x(t))
[(3
1t
tl
Ix'(s)If'(x(s)) ( )d ]'
J2(x(s)) V S S
Ix'(t)If'(x(t)) ()
J2(x(t)) v t .
(4.4.13)
On the other hand a direct differentiation gives
v(t) )'
f(x(t))
f(x(t))
v'(t)
f(x(t))
_ v(t)f'(x(t))v'(t)
J2(x(t))
v(t)lx'(t)If'(x(t))
+
J2(x(t))
(4.4.14)
344
Cha.pter 4
Equating (4.4.13) and (4.4.14), we get v'(t) = 0 for t E [tb bj and

so, v(t) = V(tl) = f3f(x(iJ)). Hence, v*(t) = f3f(x(t 1)). The inequality
(4.4.11) reduces to (4.4.5).
Ca.se 2. Suppose x(t) is negative. Then, (4.4.8) implies that w(t) > 0,
or x'(t) > 0 for t E [tl,bj. With u(t) = w(t) = a(t)[x'(t)['>' t E [tl,bj
relation (4.4.8) yields
u(t)
-f3f(x(t))
it[-
t,
-f3f(x(t)
f(x(t))] [x'(s][f'(x(s)) u(s)ds

f (x(s)
1,t K(t,s,u(s)ds,
(4.4.15)
t E [tl,b]
where K(t, s, y) = -K(t, s, y).

We note that for fixed t, s, K(t, s,) is nondecreasing. Once again
applying Lemma 4.4.1 to (4.4.15) with p(t) = -f3f(x(t gives (4.4.11)
where v*(t) is the minimal solution of the equation
v(t)
-f3f(x(t
-f3f(x(t
it
i
K(t, s, v(sds
t,
t
t, [- f(x(t]
[x'(s)[f'(x(s
j2(x(s
v(s)ds.
(4.4.16)
Now, employing an argument similar to that of in Case I, we obtain v'(t) =

and hence v*(t) = V(tl) = -f3f(x(td). The inequality (4.4.5) is now
immediate from (4.4.11).
Lemma 4.4.3. Suppose that
T/(t) =
Jto
a-1/0I.(s)ds
-00
and
<
and
lim T/(t)
t---+oo
roo q(s)ds
Jto
lim [f(x)[
Ix 1---+00
<
00,
(4.4.17)
(4.4.18)
00
(4.4.19)
00.
If x(t) is a nonoscillatory solution of equation (4.4.1), then
00
to
a(s)
.
[x'(s)["+l f'(x(s
j2(x(s
ds <
t~~ a(t)
[x'(t)["-lX'(t)
f(x(t
= 0
00,
(4.4.20)
(4.4.21)
345
and
a(t)
lx' (t) la-l x' (t)

f(x(t))
00
a(s)
. t
Ix'(s)la+l f'(x(s))
J2(x(s))
ds
[00 q(s)ds
t
(4.4.22)
for all sufficiently lage t.
Proof. Without loss of generality let x(t) be an eventually positive

solution of (4.4.1). First, we shall show (4.4.20). Suppose (4.4.20) does not
hold. Then from (4.4.18) we find that (4.4.4) is satisfied for all t 2 tt,
where tl is sufficiently large. Apply Lemma 4.4.2 to conclude that x'(t) <
o for t 2': it. Thus, (4.4.5) provides a(t)lx'(t)l<> 2': f3f(x(t l )) for t 2': it,
or
x'(t) :::.; - [f3f(x(tdW/oa- 1 /o(t) for t 2': tl'
(4.4.23)
Integrating (4.4.23) from tr to t, we get
x(t) :::.; x(tr) - [f3f(X(tl))]l/a
[t a-l/a(s)ds --+
as
(X)
t --+
(X),
t,
which is a contradiction to the fact that x(t) > 0 eventually.

Next, we shall verify (4.4.21) and (4.4.22). Integration of (4.4.7) from
to to t gives
a(t)
Ix'(t)la-lX'(t)
f(x(t))
( ) Ix'(to)l<>-lX'(tO)
a to
f(x(to))
-
to
it (
to
)d
IX'(s)l<>+l f'(x(s))
a(s)
J2(x(s))
ds.
(4.4.24)
In view of (4.4.18) and (4.4.20) it is clear from (4.4.24) that the limit
L
exists. Now, let t --+
a(t)
(X)
t~~ a(t)
Ix'(t)la-lx'(t)
f(x(t))
(4.4.25)
in (4.4.24) and then change to to t, to obtain
Ix'(t)la-lX'(t) _
f(x(t))
- L+
[00 q(s)ds + [00 a(s) Ix'(s)la+l

f'(x(s))
J2(x(s))
ds.
t
(4.4.26)
Therefore, (4.4.21) and (4.4.22) hold if L = O. For this, we consider the
following two cases:
Case 1. Suppose L < O. Then, (4.4.18) and (4.4.20) lead to
L
3
for
t 2 tl
(4.4.27)
Chapter 4
346
a(s)
Next, substituting t
= to
and
00
1 t,
a to
L
3
Ix'(s)I<>+1 f'(x(s)) I
J2(x(s))
ds <
(4.4.28)
in (4.4.26), we get
Ix'(to)l"-lX'(tO)
f (x (to) )
-1
00
to
(
)d
qS S
-1
00
to
as
Ix'(s)lo+ l f'(X(S))d
J2 (x (s ))
s.
(4.4.29)
Now, using (4.4.29) and the inequalities (4.4.27) and (4.4.28), we find for
t 2: tl,
-a(to)
Ix'(to)I<>-lX'(tO)
f(x(to))
_ L
-1
00
q(s)ds
2: - L
lt
-1
to
q(s)ds +
00
t,
ltl
to
a(s)
Ix'(s)I,,+l f'(x(s))
J2(x(s))
ds
a(s) IX'(s)I;+l f'(x(s)) ds

f (x(s))
+ "3 +"3 = -"3 ==
(3
> 0,
i.e., (4.4.4) holds. Hence, we can apply Lemma 4.4.2 to obtain a contradiction as in the proof of (4.4.20).
Case 2. Suppose L > O. From the definition of L in (4.4.25), we can
assume that
L
Ix'(t)I<>-lX'(t)
a(t)
> - for t > t .
(4.4.30)
f(x(t))
From (4.4.30), we find that x'(t) > 0 for t 2: tl. Now, using (4.4.30) in
(4.4.20) leads to
00
>
00
t,
a(s)
Ix'(s) 1<>+1 f'(x(s))

L
J2(x(s))
ds 2: "2
00
t,
x'(s)f'(x(s))
f(x(s)) ds
~ lim In [ f(x(t)) ]
2 Hoo
f(X(tl)) '
from which it follows that limHoo In[f(x(t))] <
lim f(x(t)) <
00,
or
00.
t-400
(4.4.31)
In view of (4.4.19) and (4.4.31), x(t) must be bounded.

On the other hand in view of the monotonicity of f and x, relation
(4.4.30) gives a(t)lx'(t)l<> 2: (L/2)f(x(t)) 2: (L/2)f(x(h)) for t 2: tl, or
347
Integrating this inequality from tl to t and then using (4.4.18), we find

that x(t) is unbounded, which is a contradiction. This completes the
proof.
Theorem 4.4.1. Let condition (4.4.17) hold,

00
/
du
jI/e>(u) <
and suppose Q(t)
= Loo q(s)ds
exists and Q(t) ~ 0 for t ~ to. If either
(1a(s) Q(s) )
00
l/e>
( 4.4.32)
<00
and
00
ds =
( 4.4.33)
00,
or
> k > 0 for x =I- 0 and k is a constant
f'(x)
)/
f (e>-l
e>(x)
(4.4.34)
and
(4.4.35)
00,
or condition (4.4.34) and

lim sup
Hoo
t (_1_
ito a(s)
[Q(s)
+k
00
a-l/e> (u)Q(e>+l)/<> (U)dU] )
1/0
ds =
00,
( 4.4.36)

Proof. Suppose equation (4.4.1) has an eventually positive solution x(t).
Then by Lemma 4.4.3, (4.4.22) holds. Thus, we have
1
Ix'(t)Ie>-lx'(t) > -Q(t)
>
f(x(t
- a(t)
-
and hence x'(t)
> 0 for t
x'(t)
jI/a(x(t
it follows that
X (t)
x(tll
du
jI/e>(u) >
it
tl
for
t>t
- 1
for some
tl >_ to,
tl. Now, since
(1a(t) Q(t) )l/a

(1a(s) Q( s) )
l/a
ds --+
00
as
t --+
00.
Chapter 4
348
This contradicts (4.4.32). We also find that
f'(x(s)) ds
j OOa(s)(x'(s))O+l
J2(x(s))
= jOOa(s)(x'(S))O+l (
f'(x(s))
) ds
f(O+ll/o(x(s)) f(O-ll/o(x(s))
00
a-l/O(s)Q(o+ll/<>(s)ds.
This contradicts (4.4.20) if (4.4.35) holds. Finally, it is easy to check from

(4.4.22) that
a(t)(x'(t))'" > Q(t)

f(x(t))
+k
00
a-l/"'(s)Q(a+ll/o(s)ds,
t ~ tl
or
~ tl .
x'(t)
> (_1_ [Q(t) + k jOOa-l/"'(s)Q(a+ll/"'(S)dS]) 1/", for t
J1/"'(x(t)) - a(t)
t
Integration of the above inequality from tl to t in view of (4.4.32) leads

to a contradiction. This completes the proof.
For subsequent results, we let
ho(t) =
00
q(s)ds
exist ?nd
ho(t)
~0
for all large t
and for a positive integer n and a positive constant k define the following
functions
and
00
a-l/",(s) [ho(s)
+ khn(s)]("'+ll/'" ds,
= 1,2,.
Theorem 4.4.2. Let conditions (4.4.17) - (4.4.19) and (4.4.32) hold. If

for every constant k > 0 there exists an integer N such that
hn(t) exists for n=0,1,,N-1 and hN(t) does not exist, (4.4.37)
Proof. Suppose equation (4.4.1) has an eventually positive solution x(t).
Then from (4.4.22), we have
a(t)lx'(t)I"'-lx'(t)
f(x(t))
roo a(s)lx'(s)la+l
ho(t)
+ it
ho(t)
f(a+ll/a(x(s))
for
tl
f'(x(s))
j<a-ll/a(x(s))
for some
tl
ds
to.
(4.4.38)

Clearly,
t
:> tl'
349
x'(t) ~ 0 for t ~ tl and x'(t) ~ [ho(t)f(x(t))/a(t)F/<>

As in Theorem 4.4.1, we find
))<>+1
100
it/00 a(s) ('(
;2tC(S)) f'(x(s))ds ~ k t a-I/<>(s)[ho(s)]C<>+I)/<>ds =
for
khl(t).
(4.4.39)
If N = 1 in (4.4.37), then the right-hand side of (4.4.39) does not exist.
This is a contradiction to (4.4.20).
Next, it follows from (4.4.38) and (4.4.39) that
a (t ) Ix' (t ) Ic> - 1 x' (t )
f(x(t))
~ ho(t)
+ kh1(t)
for
t ~ tl
(4.4.40)
Using a similar technique and relation (4.4.32), we get for t
l Xl
[00ka-c;1(s) [ho(s)+khl(S)] a+l ds = kh2(t).
Ix'(s)Ic>+l
a(s) J2(x(s)) f'(x(s))ds ~.
~ tl ,
(4.4.41 )
If N = 2 in (4.4.37), then the right-hand side of (4.4.41) does not exist.
This again contradicts (4.4.20). A similar argument yields a contradiction
for any integer N > 2. This completes the proof.
Remark 4.4.1. Lemmas 4.4.2 and 4.4.3 and Theorems 4.4.1 and 4.4.2
can be obtained for forced equations of the form
(a(t)lx'(t)IO-Ix'(t))'
+ q(t)f(x(t)) =
e(t),
(4.4.42)
roo
le(s)lds < 00. In this case the conwhere e(t) E C([to, (0), JR) and
clusion of Theorems 4.4.1 and 4.4.2 need to be replaced by: every solution
x(t) of equation (4.4.42) is either oscillatory, or liminfHoo Ix(t)1 = o.
The formulation of these results are left to the reader.
To prove our next linearization result, we need the following lemma.
Let a > 0 be a constant, a(t) E C([to, (0), JR+) and
C([to,oo),JR). Then the half-linear differential equation
Lemma 4.4.4.
q(t)
(a(t)lx'(t)I,,-lX'(t))'
+ q(t)lx(t)I<>-lX(t)
is nonoscillatory if and only if there exist a number T

w(t) E C 1 ([T, (0), JR) such that
w'(t)
+ q(t) + aa- 1/<>(t)lw(t)IC<>+1)/<> s=
=
~
0 on
to and a function
[T, (0).
Next, we shall consider a special case of (4.4.1), namely, the equation
(a(t)lx'(t)I,,-IX'(t))'
+ q(t)lx(t)Ii3- 1 X(t) =
0,
(4.4.43)
Chapter 4
350
where {3 > 0 is a constant.
> 0: and condition (4.4.17) hold, and
Theorem 4.4.3. Let (3
liminf
t-+co
it
to
q(s)ds > -
(4.4.44)
00.
If for every positive constant c the half-linear differential equation
(a(t)1]Q(t)ly'(t)I Q- 1 y'(t))' + cq(t)ly(t)IQ-ly(t) = 0
(4.4.45)
is oscillatory, then (4.4.43) is oscillatory.

x(t) > 0 for t?:: to. Define wet) = a(t)lx'(t)IQ- 1 x'(t)/x i3 (t) for t?:: to.
Then for t?:: to, we have
Proof.
w'(t)
-q(t) - (3a(t)
Ix'(t)IQ+l
xi3+ 1 (t)
-q(t) - {3a- 1/ Q(t)lw(t)I(a+I)/Q (x(i3- )/a(t)),
(4.4.46)
and hence
wet) = w(to) -
to
q(s)ds - /3
it
to
a(s)
Ix'(sW+ 1
X
i3+ 1 ( ) ds.
(4.4.4 7)
I(to, t)
to
a(s)
IX'(SW+l
X
i3+l() ds
S
there are two cases to consider:

Case 1. I( to, 00) < 00. In this case there exists a positive constant N
such that I(to, t) ~ N for t?:: to. For t?:: to we use Holder's inequality,
to obtain
x1-'Y(t) - x1-'Y(to)
b-
~ b-1)(I(to,t))<>~,
1)
(l:
1: :~~;)
I
Ids
a- i (S)dS) <>+1
where 'Y = ({3 + 1)/(0: + 1) > 1.

By condition (4.4.17) there exist a constant M > 0 and T?:: to such
that Xl-'Y(t) ~ M1]Q/(a+l)(t), or x(t)?:: (1]-a/(a+l)(t)/M)1/b- 1 ) for
t ?:: T and
(4.4.48)

where
Cl
351
= (1/M)(l3-a)/la h -l)].
Using (4.4.48) in (4.4.46), we find
W'(t) -S
~ q(t) ~
(c 1 /3)
0' (
0'
a(t)rJ""(t)
1/",
Iw(t)I(",+I)/a
for
~ T.
But, now in view of Lemma 4.4.4 equation (4.4.45) is nonoscillatory, which

is a contradiction.
Case 2. I(t o , 00) = 00. From condition (4.4.44) in (4.4.47) it follows that
for some constant >..,
~w(t)
2': >.. + /3I(to, t)
for every
t 2': to.
(4.4.49)
Let T 2': to be snch that A = >.. + f3I(to, T) > O. Then, (4.4.49) implies
that w(t) is negative on [T,oo). Now, (4.4.49) gives
Ix'(t)I
+1
/3a(t) x!3+ 1 (t)

>.. + /3I(to, t)
> ~ /3 X' (t)
x(t) ,
t 2': T
and consequently for all t 2': T,
In [>.. + /3~(to, t)]
> In (X(T))f3
~
x(t)
Hence, >.. + /3I(to, t) 2': A (x(T)/x(t))f3. Therefore, (4.4.49) yields x'(t)::;

~Ala-l/"'(t) for all t 2': T, where Al = (Ax f3 (T))l/a > O. Thus, we have
x(t) -S x(T) ~ Al J;a-1/a(s)ds for all t 2': T, which in view of (4.4.17)
leads to the contradiction limt-tcx) x( t) = ~oo.
Now, for (4.4.1) we shall prove the following linearization result.

Theorem 4.4.4. Let conditions (4.4.17) and (4.4.34) hold. If the halflinear differential equation
0,
(4.4.50)

Proof. Suppose equation (4.4.1) has an eventually positive solution x(t),
say, x(t) > 0 for t'2 to. Define w(t) = a(t)lx l (t)I"-lx' (t)/f(x(t)), t 2': to.
Then for t 2': to, we have
W'(t) =
~ q(t) ~ a- 1 /a(t)lw(t)l(a+l)/a ( f (a-f~~~(t))

))).
a(x(t
(4.4.51)
352
Chapter 4
Using condition (4.4.34), we get
~(a)a-1/<>(t)lw(t)I(0+1)/0
for t 2': to.

a
Now, applying Lemma 4.4.4 we find that (4.4.50) is nonoscillatory, which
is a contradiction.
w'(t) :-:; - q(t) -
To relax condition (4.4.34) from Theorem 4.4.4 and to extend Theorem

4.4.3 to more general equations of type (4.4.1), we require the following
condition:
00
(F).[
(f'(u)) 1/(0+1)
J2(u)
du <
-00
00
and
(f'(U)) 1/(<>+1)
J2(u)
du <
00
(4.4.52)
and
. {.
f'(x)
;~~ j(o-l)/o(x)
mm
inf (
x<o
) 0/(0+1)
roo (f'(u)) 1/(<>+1)
ix
1-
f'(x)
)0/(0+1)
j(o-l)/o(X)
x
Rem.ark 4.4.2. For a
J2(u)
00
du,
(f'(u) )1/(0+1) dU} > O.

J2(u)
= 1 conditions (F) and (F 2 ) are the same.
Theorem. 4.4.5. Suppose conditions (F), (4.4.17) and (4.4.44) hold. If

for every constant C1 > 0 the equation
+ c1q(t)lz(t)I<>-l z (t)
(a(t)rt(t)lz'(t)IO- 1z'(t))'
(4.4.53)

x(t) > 0 for t 2': to. Define wet) as in Theorem 4.4.4 and obtain (4.4.51).
Then, we have wet) = w(to) - It: q(s)ds - :r(to, t), where
Proof.
:r(to, t)
t
to
Ix'(s)lo+ 1 ,
a(s) J2(x(s)) j (x(s))ds.
There are two cases to consider:

Case 1. :r (to, 00) < 00. In this case there exists a positive constant N
such that :r(to, t) :-:; N for t 2': to. For t 2': to, we use Holder's inequality
to obtain
li
to
x'es)
(f'(X(s)))
J2(x(s))
O~,
ds
< (:r(to, t)) Q~'
(i a- 1/O(s)ds
t
to
< N 1/(<>+1)'I]0/(0+1)(t) for
0+1
t 2': to.
(4.4.54)
353
Condition (F) ensures that
f'(x(t))
)<>/(0+1) (= (f'(u) )1/(0+1) du 2 5
f(o-I)/Oo(x(t))
.Jx(t)
J2(u)
for
(4.4.55)
where 5 is a positive constant. Let
1=
* -
t 2 to,
(f'(v.))I/(Oo+1)
~()
d11. > O.
x(to)
Then in view of (4.4.55) for t 2 to, we obtain
f'(x(t))
f(Oo-l)/Oo(X(t))
5(0+1)/0
(f'(
)) du
_11._
J2(11.)
[K* _ l (f'(_11._))
[1
1/(a+1)
]-(<>+1)/<'>
x(t)
5(00+1)/0
x(to)
L
t
5(0+1)/<>
1/(0+1)
X(t)
K* -
2 5(0+1)/0 [K*
x'(s)
lit
to
d11.
] -(00+1)/"
J2(u)
,
(;2~:~:~D
]-(0+1)/<;
1/(<>+1)
ds
x'(s) (f'(x(s))) 1/(0+1) dsl]-(O+l)/Oo

J2(x(s))
and hence on using (4.4.54), for every t 2 to, we get
and consequently, in view of (4.4.17), we find
f'(x(t))
C
jCa-l)/a(x(t)) 2 7)(t)
for all
(4.4.56)
t 2 to,
where c is a positive constant which depends on 5, K*, Nand a.

Now, from (4.4.56) equation (4.4.51) gives
w'(t)
:<::::
q(t) - a
(~a (a(t)7)a(t))l/o
1
)
Iw(t)l(a+1)/Oo
for
t 2 to.
354
Chapter 4
Case 2. J (to, (0) = 00. The proof for this case is similar to that of Theorem
4.4.3 Case 2 and hence omitted.
Remark 4.4.3.
1. If
1, then Theorems 4.4.4 and 4.3.5 are the same.
2. From Theorems 4.4.3 ~ 4.4.5 it is clear that the oscillatory results for
equations presented in Chapter 3 can be used to obtain interesting oscillation criteria for nonlinear equations of type (4.4.1). The
formulations of these results are left to the reader.
half~linear
3. Theorems 4.4.1 and 4.4.2 hold for (4.4.1) without (4.4.34). This can
be proved by using Theorem 4.4.5 and replacing tpe function a(t) with
a(t)rl"'(t) in (4.4.35) and (4.4.36) of Theorem 4.4.1 and in hl(t) and hn(t)
in Theorem 4.4.2. Further, in both theorems k > 0 is any constant.
In the following results we shall obtain oscillation criteria for (3.13.76)
and (3.13.89) when g(t) = t.
Definition 4.4.1. We say that (3.13.76) is strongly superlinear if there

exist a constant c > 0 such that lyl-CIF(t, y)1 is nondecreasing in Iyl
for each fixed t, and
(4.4.57)
In view of this definition (4.4.43) is strongly superlinear if (3 >
Now, we shall consider (3.13.76) with g(t)
10
00
1'1/1-1
(a~s)) 1ds
00
Q.
t and assume that
for every constant
ki= o.
Theorem 4.4.6. Suppose (3.13.76) is strongly superlinear and

(4.4.58)
Equation (4.4.76) is oscillatory if and only if
00
for every constant k
(4.4.59)
i= O.
Proof. The 'only if' part follows from Theorem 3.13.12. To prove the
'if' part assume that (3.13.76) has a nonoscillatory solution x(t), say,
x(t) > 0 for t ~ to. Integrating (3.13.76) from t to T and letting
355
T --t 00 and noting that limt->oo a(t)ljJ(X'(t))

ftOO F(s, x(s))ds, which implies
x'(t)
1p-1 (att)
1=
~ 0,
we find a(t)1jJ(x'(t)) ~
( 4.4.60)
F(s, X(S))dS) .
We divide the above inequality by 1j;-1(X C(t)), where c > 0 is a consta.nt

of strong superlinearity of equation (3.13.76) and use (4.4.58), to obtain
x' (t)
> 'Ij; -1
> 1j;-1
-1
a(t)
F(s,x(S))d .5 )
00
xC(s)
(_1_
roo F(s, x(s)) dS)
a(t) it
XC(S)
for
to
Since :r(t) ~ m for some constant m > 0, ill view of the strong superlinearity of (3.13.76), we have x-C(t)F(t,x(t)) ~ m-CF(t,m) for t ~ to,
so that
x'(t)
((m-
00
>
1p-1
>
1jJ-1(m- C)1jJ-1 (att)
att)
F(s, m)ds)
Loa F(s, m)ds)
for
to.
Integrating tIle last inequality from to to t, we obtain
which because of (4.4.57) implies
Loa 1jJ-1 (a(~) 1= F(u,m)du) ds
<
00.
But this contradicts (4.4.59) and completes the proof.
The following result is a variant of Theorem 4.4.6.

Theorem 4.4.7. Suppose condition (4.4.58) holds and there exist functions q(t) E C([t o,oo),1R+) and 9 E C(lll+,1R+) such that 9 is increasing,
(4.4.61 )
IF(t, y)1 ~ q(t)g(IYI) for (t, y) E lR+ X lR
and
00
du
<
1j;-l og (U)
00
for any
> O.
( 4.4.62)

( 4.4.63)
Chapter 4
356
Proof. Since the 'only if' part follows from Theorem 3.13.12, it suffices
to prove the 'if' part. Assume that (3.13.76) has a nonoscillatory solution
x( t) > 0 for t ~ to > O. As in Theorem 4.4.6, we obtain (4.4.60) which
implies
x'(t)
1p-1 (a(t)
00
q(s)g(X(S))dS) ,
~ to
Dividing the above inequality by 1p-1(g(x(t))) and usiug (4.4.58), we get
x'(t)
. -1 (
> 1/J
>
1/J-l
;;'-(0
(a(t)
j'CXJ
t
g(x(s)))
q(s) g(x(t)) ds
[X) q(S)dS)
for
(4.4.64)
~ to,
where we have used the fact that the function g( x( s)) is increasing and
~ 1 for s ~ t.
g(x(s))/g(x(t))
Integrating (4.4.64) from to to t, we find
l t (1 1
00
1/J-l
to
a(s)
q(u)du ) ds:S:
which in view of (4.4.62) implies
L=
1/J-1
(ats)
lt
x'(s)
ds =
1/J-1(g(x(s)))
to
00
q(U)dU) ds <
(t)
x(tol
1p-1
dw
,
0 g(w)
00.
But this contradicts condition (4.4.63) and completes the proof.
ExaIllple 4.4.1. Consider (4.4.43) and assume that ICXJ a-1/n(s)ds = 00.
From Theorem 4.4.7 it is clear that a necessary and sufficient condition for
(4.4.43) to be oscillatory is that
00
to
Jt
OO
a(s)
[= q(u)du)
1/0.
ds
00
if
(3 > a.
Next, we shall consider equation (3.13.89) with g(t)

a-l/n(s)ds and assume that 7r(to) < 00.
= t. Let 7r(t)
TheoreIll 4.4.8. If for every nonzero constant k,
(00 IF(s, k7r(s))lds
lto
and
(00
lto
( 4.4.65)
00
(1a(s) ltot IF(u, k7r(u))ldu )1/'" ds
00,
(4.4.66)
357

Proof. In view of Theorem 3.13.17 condition (4.4.65) ensures the nonexistence of a nonoscillatory solution of type III of equation (3.13.89), (i.e.,
limHoo Ix(t)11 7T(t) constant> 0). Since (4.4.66) implies
the equation (3.13.89) has no nonoscillatory solution of type (I), (i.e.,

limt-+oo Ix(t)1 = constant> 0). Thus, it suffices to show that there is no
nonoscillatory solution of type II of (3.13.89), (i.e., limt-+oo x(t) = 0 and
limHoo Ix(t)I/7T(t) = (0). Let x(t) be a type II solution of (3.13.89) which
is eventually positive. Then, x'(t) is eventually negative and proceeding
as in Theorem 3.13.16, we obtain
Since x(t) satisfies C27T(t):::; x(t) for t

the above inequality implies that
to for some constant
C2
> 0,
which contradicts the condition (4.4.66). Similarly, (3.13.89) does not admit
eventually negative solutions of type II. This completes the proof.
Theorem 4.4.9. Let equation (3.13.89) be strongly superlinear. Then

equation (3.13.89) is oscillatory if and only if
{DO IF(s, h{s))lds
lto
00
(4.4.67)
for any nonzero constant k.

Proof. The 'only if' part follows from Theorem 3.13.17. To prove the
'if' part, assume that (3.13.89) has a nonoscillatory solution x(t) > 0 for
t ~ to ~ 0. Then either
(i)
x'(t) > 0, t
to,
or
(ii) x'{t) < 0, t ~ tl for some tl ~ to.
If x'(t) > 0 for t ~ to, then x(t) ~ Cl, t ~ to for some constant
> 0. As in Theorem 3.13.16, we find that
F(s,x(s))ds < 00 and
hence
F(s, cl)ds < 00. This implies that
F{s, Cl7T(S))ds < 00,
(i)
Cl
It:'
It:'
It:'
Chapter 4
358
(ii) If x'(t) < 0 for t 2': tl, then there exists a constant C2 > 0 such that
x(t} 2': C21T(t) for t 2': t l , (cf. Lemma 3.13.9), and strong superliniearity
of (3.13.89) implies that
(4.4.68)
where c >
is given in the Definition 4.4.1.
Next, as in Lemma 3.13.9, we obtain
x(t) 2': - (a l / (t))1T(t)X'(t)

Q
for
t 2': t l
Using (4.4.68) and (4.4.69), we find by noting (x')"" =
[_ ( -a(t)(x'(t))"*r(C-O)/"], =
C: 0(-a(t)(x'(t))"*)
-c/o
c:
Ix'l"
(4.4.69)
sgn x' that
(-a(t)(x'(t))""r c / o F(t, x(t))
(XC(t)) (x-C(t)) F(t, x(t))
2': c:o (-a(t)(x'(t))o") -;'; [-a(t)(x'(t))o*J;'; (1TC(t))(c21T(t))-CF(t,c21T(t))

c-o
o
--ciCF(t,c21T(t))
for
t2':t l .
An integration of the above inequality yields

co
--cic
it
t,
F(s, C21T(S))ds:S
-a(td(x'(td)O
which implies that It"': F( s, C21T( s) )ds <

dition (4.4.67).
00.
"J -(c-a)/o
This again contradicts con-
Example 4.4.2. Once again consider (4.4.43) and assume that aCt) = eAt
and q(t) = eJ.Lt, t 2': 0 where ), and J-t are constants and ), > O. Let
o < (3. Then by Theorem 4.4.9 equation (4.4.43) is oscillatory if and only
if J-t/), 2': (3/0.
4.5. Superlinear Forced Oscillations

Here, we shall discuss the oscillatory behavior of the equation
(a(t)x'(t))'
+ p(t)x'(t) + q(t)lx(t)I'"Y
sgn x(t)
= e(t),
(4.5.1)
where I> 0, aCt), e(t), pet) and q(t) E C([to,oo),lR) and aCt) > 0 for
t 2': to
359
Theorem 4.5.1. Suppose for every T 2 0 there exist T:'S Sl < t1 :'S
S2 < t2 such that
e(t) { :'S 0 for

20 for
If there exist p(t) E C 1 ([to,oo),lR+)
C 1([Si' til, JR): u(t) # 0, u(s;) = u(t i )
constant c> 0,
Qi(U) =
for 'i
lSi(i p(s) [q(s)u 2(s) -
E
t E
[Sl' tIl
[S2, t2].
and
u(t)
(4.5.2)
E
D(Si,ti)
{u(t)
= O},i = 1,2 such that for every
(3()
ca.(s) (u.'(s)
+ a(s);(s) u(s)
)2j ds
2 0
(4.5.3)
= 1,2 where (3(t) = a(t)p'(t) - p(t)p(t), then
(i) every unbounded solution of equation (4.5.1) with I> 1 is oscillatory,

(ii) every bounded solution of equation (4.5.1) with I < 1 is oscillatory.
Proof. Suppose x(t) is a nonoscillatory solution of equation (4.5.1), say,
x(t) > 0 for t 2 to 2 o. Define w(t) = -p(t)a(t)x'(t)/x,(t) for t 2 to.
Then for t 2 to, we have
,_
_ e(t)p(t)
w (t) - p(t)q(t)
()
x' t
+ IX
,-I
w 2 (t)
j3(t)
(t) I) ( ) + () ( ) lL(t).
a, t p t a t p t
I
(4.5.4)
By the hypothesis we can choose Sl, t1 2 to so that e(t):'S 0 on the

interval [131, tIl = I with Sl < h. On the interval I, w(t) satisfies (4.5.4)
and the inequality
w'(t) 2 p(t)q(t)
+ j3(t) w(t) + , X,-l(t) u,2(t)

a(t)p(t)
on
a(t)p(t)
I.
(4.5.5)
Let u(t) E D(Sl' t1) be as in the hypothesis. Multiplying (4.5.5) by u 2(t)

and integrating over I, we find
Jf 2u(s)u'(s)w(s)ds
(3(s)
Jf u 2(s)p(s)q(s)ds + . f a(s)p(s)
u 2(s)w(s)ds
(
+ Jf ,X
1
2
w 2 (s)
- (s)u (s) a(s)p(s)ds,
which is equivalent to
02 (
Jf
[(IX'-l(S))~ u(s)w(s) + (u'(s) +

a(s)p(s)
( [p(s)q(s)u2(s) -
+ Jf
a(s)p(s) ( ,
IX,-l(S) U (s)
u(s){3(s))
2a(s)p(s)
(a(s)p(s))~] ds
u(s){3(s)
+ 2a(s)p(s)
IX,-l(S)
)2] ds.
(4.5.6)
360
Chapter 4
(i) If x(t) is an unbounded nonoscillatory solution of equation (4.5.1)

with 'Y > 1 on I, then there exists a constant Kl > 0 such that
x(t) ~ Kl on I. Then, we have
1
K 1- ,
_1_
'Y
'Y
-xl-,(t) <_
CIon
I,
(4.5.7)
where Cl is a constant. Using (4.5.7) in (4.5.6) and proceeding as in

Theorem 2.7.4, we arrive at the desired contradiction.
(ii) If x(t) is a bounded nonoscillatory solution of equation (4.5.1) with
'Y < 1 on I, then there exists a constant K2 > 0 such that x(t)::; K2
on I, and hence (l/'Y)xl-,(t)::; Ki-' /'Y = C2 on I, where C2 is
a constant. The rest of the proof is similar to that of the above case and
hence omitted.
Next, we shall present the following criterion for the oscillation of all
solutions of the superlinear equation (4.5.1), i.e., when 'Y> 1.
Theorem 4.5.2. Suppose for every T ~ 0 there exist T::; Sl < tJ ::;
S2 < t2 such that condition (4.5.2) holds and q(t) ~ 0 on [Sl, tl]U[S2' t2].
If there exist pet) E Cl([to, 00), JR+) and v,(t) E D(Si' ti), i = 1,2 where
D(Si, t i ) is as in Theorem 4.5.1 such that
Qi(V,)
l~i [E(S)V,2(S) -
a(s)p(s) (v,'(s)
+ v,(s) 2afs\~(s) 2]
ds
(4.5.8)
for i = 1,2 where (J(t) = a(t)p'(t) - p(t)p(t) and
E(t) = (p(t)q(tlh(le(t)lp(t(r-l)h,
then equation (4.5.1) with 'Y
> 1 is oscillatory.

~ to ~ O. Define wet) = -p(t)a(t)x'(t)/x(t) for t ~ to.
Then for t ~ to, we have
Proof.
x(t) > 0 for t

,_
W
(t) - p(t)q(t)x
,-1
(J(t)
e(t)p(t)
(t) + a(t)p(t) wet) - x(t)
+ a(t)p(t) w
(t). (4.5.9)
By the hypothesis, we can choose Sl, tl ~ to so that e(t)::; 0 on the

interval [SlJ tl] = I. On I, wet) satisfies the equation
w'(t)
p(t)q(t)X,-I(t) + le(t)lp(t)
x(t)
(J(t) wet)
a(t)p(t)
a(t)p(t)
w2(t).
361
Oscilla.tion theory for superlinear diHerentia.l equations

Since, for tEl,
inf [P(t)q(t)X"l-l(t)
x>o
+ le(t)IP(t)]
2: (p(t)q(t))lh(le(t)lp(t))h-1lh
x(t)
on I, we have
1O'(t) 2: (p(t)q(t))lh(le(t)lp(t))h-1lh
=
E(t)
(3(t)
a(t)p(t) w(t)
+ (3(t) 1O(t) + ----,W-,-,2-"-;(t)=-:a(t)p(t)
a(t)p(t)
10 2(t)
a(t)p(t)
(4.5.10)
Multiplying both sides of (4.5.10) by u 2 (t) and integrating over I, we
obtain
02: {E(s)u 2(s)ds
if
([(2U I (S)U(s)
if
+ (3(S)U 2(S))
a(s)p(s)
1o(s)
+ u 2(S)10 2(s)]
a(s)p(s)
ds
= Q1(U) + ( [U(S)1O(S) + (ul(s) + (3\s\u\s\) Ja(s)p(s)] ds.
if
Ja(s)p(s)
2a s p s

If p(t) = exp
Remark 4.5.1.
(t p(s)/a(s)ds) ,
then (3(t) =
o.
Thus,
with this choice of the weight function p(t) simpler forms of Theorems
4.5.1 and 4.5.2 can be reformulated. The details are left to the reader.
Example 4.5.1.
Consider the superlinear equation
+ t12X'(t) + t"l
m sint Ix(t}l"l sgn x(t)
( ~X'(t))'
t
cost,
t > 0 (4.5.11)
where 'Y > 1 and m > O. For any T > 0 choose n sufficiently large so
that 2m!" 2: T, and let Sl = 2m!" and t1 = 2m!" + (7r /2) in (4.5.8), take
p(t) = t and u(t) = sin 7r(t - Sl)/(t1 - Sl) if t E [Sl, tlJ. Then, we have
Q1(U) =
where
1:'
[(p(s)q(s)?h(le(s)lp(s))h- 1lhu 2(s) - a(s)p(s)(u l(s))2] ds
("/2
4m 1h io
2(311"1
is the Gamma function. If
sin 2 +(lhl s
COS 3 -(lhl
(3 + 1.) r (4 - 1.)
"I
r(7)
11"1
m
"I
cos 2 2sds
7r,
7rr(7)
2:
("/2
s ds - 4 io
2r (3 + ~ ) r (4 _ ~ )'
(4.5.12)
Chapter 4
362
then Ql(n) 2: O. Similarly, for 82 = 211:rr + (7r/2) and t2 = 2n7r + 7r,

we can show that Q2(n) 2: O. It follows from Theorem 4.5.2 that equation
(4.5.11) is oscillatory if (4.5.12) holds.
4.6. Oscillation of Nonlinear Differential

Equations with Nonlinear Damping Term
In this section we sha.ll consider the differential equation
x"(t) + p(t)g(x'(t)) + q(t)f(x(t)) = 0,
(4.0.1)
where f is a superlillear restoring force, i.e., f (x) E C 1 (lR, lR), f' (x) 2:
0, xf(x) > 0 for x =J 0, f(O) = 0 and f satisfies condition (4.1.5),
g(x) E C(lR, lR) and 9 is locally Lipschitz. In addition p(t), q(t) E
C([to, 00), lR) (locally integrable would suffice).
In what follows we shall extend the well-known result, namely, if q(t) >
0, t 2: to then all solutions of (4.1.6) oscillate if and only if ftC; sq(s)ds =
00.
4.6.1. Damping with Nonlinear GrowthOscillation Criteria

We shall need the following lemmas.
!too
Lemma 4.6.1. Let g( 0) = 0 and suppose Q( t) =

q( s )ds exists
as a finite number and p(t) E LA[to, 00), to 2: 0 where 1:S: >. :s: 00. If
x(t) is a bounded, eventually monotone solution of equation (4.6.1), then
limt---+oo x'(t) = O.
Proof. Let x (t) be posi tive and monotone increasing in [T, 00) for some
T 2: to with limt---+oo x(t) = c > 0, where c is a constant. All other
possibilities can be treated similarly.
Since x'(t) 2: 0 on [T,oo) and x(t) is bounded, we find
lim inf x'(t)
t---+oo
Suppose for some
EO
O.
(4.6.2)
EO.
(4.6.3)
> 0,
lim sup x' (t) 2:
t---+oo
Define JL by (1/>.) + (1/11) = 1. We assume that EO is sufficiently small

so that Ig(n)l:S: 1 whenever Inl:S: EO. It follows from (4.6.2) and (4.6.3)
that there are disjoint intervals [an, bnl c [T, 00) with limn---+oo an = 00
363
such that EO = x'(a n ) > x'(t) > x'(b n ) = Eo/2, whenever t E (an, bn ).
Integration of equation (4.6.1) from an to bn gives
1
-2" EO
rbn
Jan
p(s)g(x'(s))ds -
ibn
a"
q(s)f(x(s))ds.
(4.6.4)
Now, from the integration by parts and the general mean-value theorem
for integrals, we get
1~" q(s)f(x(s))ds
Q(bn)f(x(bn)) - Q(an)f(x(an))
+ 1~" Q(s)f'(x(s))x'(s)ds
Q(bn)f(x(b n )) - Q(an)f(x(a n )) + Q(cn)[f(x(bn)) - f(x(an))J,

(4.6.5)
for some Cn E (an' bn ), since f'(x(s))x'(s) 2: 0 for s 2: T. Let C be the
>' norm of p, Qn be the SUPtE[an,b"jIQ(t)1 and use (4.6.4), (4.6.5) and
Holder's inequality, to obtain the estimate Eo/2 :::: C(b n -an )1/1-'+4Qnf(c).
Since Q(t) = !too q(s)ds < 00, limn-->oo Qn = 0 and so 4Qnf(c):::: Eo/4
for n 2: N, say. This gives bn - an 2: (EO/ 4C)I-', n 2: N. But, since
x'(t) 2: Eo/2 on [an, bn ], we have
contradicting limt-->oo x(t) = c. Therefore, we must have lim SUPt-->oo x'(t)

=0.
In the remainder of this section, we shall assume that there exist constants K, b with b > 1, such that
o < g(u)
sgn u :::: Klul b
for all
u,
(4.6.6)
and let ,( u) be defined by
,(u) = {
g(u) if u~O
u
o if u = O.
It follows that , is bounded, nonnegative function of u and ,[1>(u)]

is locally integrable for any 1>(u) E C(JR,JR). We also define by {3 =
rnin{b,2} and Q(t) = !too q(s)ds if exists. For 1::::.\:::: 00, >' is the
space of .\th power integrable functions on [to, (0), to 2: 0 with the usual
norm denoted by I I>.
The next two lemmas show that certain types of nonoscillatory behavior
can be excluded by appropriate sign or integral conditions on the functions
p(t) and q(t).
Chapter 4
364
Lemma 4.6.2. (i) p(t):::": 0 for all sufficiently large t and the existence
of Q(t) :::": 0 with Q(t) rf. .c1[to, 00) imply that there is no solution x(t)
of equation (4.6.1) satisfying x(t)x'(t):::": 0 for all large t.
(ii) p(t) E .c 1/(2-i3l[to,00), q(t) :::": 0 for all sufficiently large t with
Q(t) rf. .c1[to, 00) imply that there is no bounded solution x(t) of equation
(4.6.1) satisfying x(t)x'(t):::": 0 for all large t.
Proof. Assume that x(t) is a solution of equation (4.6.1) which is positive
and increasing on [to, 00). We also suppose that to is large enough so that
all sign hypotheses hold on [to, 00). Define w(t) = x'(t)/ f(x(t)), t :::": to.
Then for t:::": to, we have w(t):::": 0 for t:::": to, and
w'(t) = - q(t) - p(th(x'(t))w(t) - f'(x(t))w 2 (t).

(i)
(4.6.7)
In this case, we have
w'(t) :::": - q(t)
for
t:::,,: to.
(4.6.8)
Integrating (4.6.8) from t to T with to::; t ::; T, we obtain w(t) :::":

w(T) +
q(s)ds :::":
q(s)ds. Thus, as T -+ 00, we get w(t):::": Q(t)
for t:::": to. Now, integrating again, we find
ItT
It
1to00 Q(s)ds
::;
100 w(s)ds
to
::;
which contradicts the hypothesis that Q(t)
100
x(tol
f(~ ) < 00,

u
rf. .c1[to, 00).
(ii) For this case, we write the equation (4.6.7) in the form
(w(t)ex p
U:
P(Sh(X'(S))dS)), = [-q(t) - f'(x(t))w 2 (t)]

x exp
(1:
p(sh(x'(S))dS) .
Integrating (4.6.9) from t to T with to::; t ::; T, we find
w(t) = w(T)(T)
where (s) = exp
w(t) :::":
[q(s)
(ItS p(uh(x'(u))du).
iT
q(s) exp
(4.6.9)
+ f'(x(s))w 2 (s)](s)ds,
This yields the inequality
(100 p(uh(X'(U))dU) .
(4.6.10)
Since x(t) is bounded, by Lemma 4.6.1, we obtain limHDO x'(t) = O.

Hence, for all sufficiently large a, we have 0::; ,(x'(a)) ::; K(x'(a))b-l ::;
365
K (x' (a) )/3- 1. Thus, for s, t :::-: to sufficiently large and Holder's inequality,
we get
lIs p(a)'(x'(a))dal <:; IS Klp(a)l(x'(a))/3-1da

<:; K
(IS Ip(a)122.eda) 2-/3 (I' x'(a)da)/3-1 <:; K(lplm) (c/3- 1)
K o,
say, where Tn = 1/(2 - (3) and c = limt-hXJ x(t). Using this estimate
in (4.6.10) and the fact that q(t) :::-: 0 for sufficiently large t, we find
'W(t) :::-: e- Ko ftT q(s)ds. The rest of the proof is similar to that of Case (i)
above.
The proof of the nonexistence of negative, decreasing solutions of equation (4.6.1) is similar.
Lemma 4.6.3. If either of the following two sets of hypotheses holds,

then there is no solution x(t) of equation (4.6.1) satisfying x(t)x'(t) <:; 0
for all large t.
(IIl
[1 + J: Ip(s)ldS]
-1
t/. 1/(b-1)[to,00),
and q(t):::-: 0 for sufficiently
large t and is not identically zero on any half-line.

(h) p(t) :::-: 0 for sufficiently large t, p(t) E 1/{2-/3)[to,00),
exists and nonnegative for sufficiently large t.
and Q(t)
Proof. We shall consider only the proof of nonexistence of positive, decreasing solutions. The proof for negative, increasing solutions is similar.
Assume that x(t) is a positive solution of equation (4.6.1), monotone decreasing on [to, (0) with to large enough so that sign hypotheses hold on
[to, (0).
(h) q(t):::-: 0 implies that x"(t) + p(t)g(x'(t)) <:; 0 for t:::-: to. Let
v(t) = -x' (t). Then, we have v' (t) + p( t)g( -v( t)) :::-: 0 and so by condition
(4.6.6) ,
(4.6.11)
The hypothesis q( t) l' 0 on any half-line is to exclude the possibility
that x(t) is eventually constant. Hence, we can assume without loss of
generality that v(to) > 0 and integrate (4.6.11) over an interval [to, t] on
which v(t) is nonvanishing, to obtain
v 1- b(tO) - v 1- b(t) :::-:
- K(b - 1)
it
Ip(s)lds.
(4.6.12)
to
From (4.6.12), we find that v(t) must remain positive and
v(t) :::-:
[v1-b(tO)
+ K(b -1)
L
t
IP(S)ldS]
-1/{b-1)
,
t:::-: to
(4.6.13)
Chapter 4
366
Now, the hypotheses imply that
+ K(b -1)
V 1 - b (tO)
however, 0 -c::
Lt]-1
Ip(s)lds
rt
L: I /(b-l) [to, 00),
ftC; v(s)ds -c:: x(to) and so (4.6.13) yields a contradiction.
(I/o
(h) Define u(t) = x'(t) exp

p(s)ry(x'(s))ds) , t :;;. to Then, u(t) -c:: 0
for t:;;. to. Arguing as in Lemma 4.6.2(ii), we obtain
(Ip(s)b(x'(s))ds -c:: Ko,
say, for all
ltD
t:;;. to.
( 4.6.14)
Therefore, if limsuPt-->oo u.(t) < 0, then limsuPt-->oo x'(t) < 0, which

is impossible for a positive function x(t). Thus, limsuPt-->CXJ u(t) = O.
Define tn = inf {t : u( t) = -1/ n }. Clearly, tn is well defined for n
sufficiently large, and the sequence {tn} increases to 00. We can write
equation (4.6.1) as
u'(t)
+ q(t) exp
(1:
p(s)ry(x'(S))dS) f(x(s)) =
u(t) - u(t n )
lt
q(s) exp
f(x(t))
lt
(1:
(4.6.15)
where t E [tn-I, t n ) and using the
Integrating (4.6.15) from t to tn,

definition of tn, we get
o>
o.
p(o}y(x'(O))dO) f(x(s))ds
q(s)(s)ds
lt
f'(x(s))x'(s)
(l
tn
q(O)(O)dO) ds,
(4.6.16)
where (s)
as
exp
(It: p(o)ry(x'(o))do) . Inequality (4.6.16) can be written

F(t) _ltn H(s)F(s)ds < 0,
(4.6.17)
where
F(t) = f(x(t))
(n
it
q(s)(s)ds
and
H(t)
f'(x(t))x'(t)
f(x(t))
:;;. O.
An application of Gronwall's lemma to (4.6.17) gives F(t) < 0 for t E

[tn-I, tn), i.e., fttn q(s)(s)ds < O. Hence, lim infr--> 00 ftT q(s)(s)ds <
O. But, since
iT
q(s)(s)ds = (t)Q(t) - (T)Q(T)
iT
Q(s)'(s)ds
367
from the hypotheses it follows that C/>, c/>', Q are all nonnegative on [to, 00)
with limT---+oo Q(T) = 0, and from (4.6.14), we find that </>(T) ::; eKo for
T 2:: to Thus, limT---+CX) It q(s)c/>(s)ds 2:: O. This contradiction completes
the proof.
Now, we are prepared to state and prove the following oscillation criteria..
Theorem 4.6.1. Let condition (4.6.6) hold. Then,
(I)
p(t), q(t) eventually positive with
[1 + Itiop(s)dsr1 (j 1/(b-1)[to,
00) and Q(t) (j l[to, 00) imply that equation (4.6.1) is oscillatory.
(II) p(t) E 1/(2-i3)[to,00), q(t) eventually positive with Q(t) (j 1 [to,oo)
imply that all bounded solutions of equation (4.6.1) oscillate.
(III) p(t) eventually positive, p(t) E 1/(2-13) [to, 00), Q(t) exists and is
eventually positive with Q(t) (j 1 [to, 00) imply that all bounded solutions
of equation (4.6.1) have oscillatory derivatives.
Proof. In (I) and (II), the positivity hypothesis on q(t) implies that
any nonoscillatory solution x(t) of equation (4.6.1) satisfies the inequality
x"(t) sgn x(t) < 0 for any sufficiently large t for which x'(t) vanishes.
This implies that any nonoscillatory solution must be monotonic. The
result now follows from the appropriate parts of Lemmas 4.6.2 and 4.6.3.
For (III) we can use the above lemmas to conclude that any bounded
nonoscillatory solution of equation (4.6.1) cannot be monotonic, which is
equivalent to the statement that its derivative oscillates.
Remark 4.6.1. The condition p(t) E 1/(2-13) [to, 00) in (II) is stronger
than the condition
[1 + Ittop(s)dsr1 (j 1/(b-1)[to,00).
4.6.2. Damping with Nonlinear GrowthA N onoscillation Theorem

We shall prove the following nonoscillation result for equation (4.6.1).
Theorem 4.6.2. Suppose the following conditions hold
(i1) g(O) = 0 and there exist constants K, b with b 2:: 1 such that for
all u, v with lui::; 1, Ivl:::; 1,
Ig(u) - g(v)1 ::; Klu -
vi [lulb-1 + Ivlb- 1 ]
(i2) q(t)El[to,OO), to 2::0 and there exist To 2:: to, k with k>l such
Chapter 4
368
that
Ql(t) =
100
q+(s)ds > k
100
q-(s)ds
(b) p(t)EL l /(2-,13)[to,00), f3=min{2,b}.

Then, ftOO Q(s)ds < 00 implies that there is a nontrivial nonoscillatory
solution of equation (4.6.1).
Proof. We note that the proof presented here is similar to that of Theorem
4.3.1. We observe that condition (i2) implies that for t ~ To,
itrOO [q+(s) -
Q(t)
>
q-(s)] ds ~
(k-1) QI(t)
-k-
k- 1) 100 [q+(s) + q-(s)] ds

(~
t
(k-k- 1) Q3(t),
where Q3(t) = !too Iq(s)lds. Thus, Q(t) E LI[to, 00) shows that Q3(t) E
.c1[to, 00). We also note that Q3(t) ---+ 0 monotonically as t ---+ 00. Let
c > 1 and define
A =
max
f(u).
(4.6.18)
c-I:O;u:O;c+1
Choose T
To so large that
1
Q3(T) < 2A'
loo
Q3(s)ds
min{1,1/(2A)}
(4.6.19)
(4.6.20)
and in view of condition (b),

(4.6.21)
Let S be the space of functions x(t) which are bounded and absolutely
continuous on [T, 00) and whose derivatives are a.e. bounded above by
some mUltiple of Q*(t) = 2AQ3(t). We make S into a Banach space by
introducing the norm Ilxll = Ilxll oo + IIQ:;-lx'lIoo, where II 1100 is the
uniform norm on [T, 00).
For n = 1,2, define Bn to be the set of functions xES which
are constant on [T + n, 00) and satisfy the inequalities
Ilx - cll oo <
1,
(4.6.22)
IIQ;lx'll <
(4.6.23)
369
and
IX'(t2) - x'(tdl :s; A

for all t l ,
+K
q(s)ds
t, -
[It?-lp(s)l l /(2-(3)ds J
2-(3
t,
(4.6.24)
[T,T+n].
t2 E
Clearly, En is a convex subset of S and by the Ascoli-Arzela theorem

En is compact. Define the mapping Y n on En by
c
(Ynx)(t) =
-100
lO[p(a}9(x'(a))+q(a)f(x(a))]dads, T:S; t:S; T+n
r r
JT+nJs
(X)
C -
00
q(a)f(x(a))dads, T
+ n < t.
Since for any function x E En, x( t) is constant on [T + n, 00), the

integrability of q(t) and Q(t) imply that Yn maps En into S. We
shall now verify that Yn(En) C En Let x E En. If T:S; t :s; T + n, we
have
(4.6.25)
Ix'(s)l:s; Q*(s) = 2AQ3(S) :s; 2AQ3(T) :s; 1, it follows that

lx' (s) Ib :s; lx' (s) 1(3. Thus, on applying Holder's inequality and the fact that
Now, since
Q * (t) is a monotonic function, we obtain
IX) Ip(s)llx'(s)lbds :s; 1 Ip(s)llx'(s)l(3ds

00
:s;
1 Ip(s)IQ~(s)ds
00
:s; Q*(t) [1OOIP(S)11/(2-(3)dSr-(3
(1
Q.(s)ds )(3-1
00
( 4.6.26)
Combining (4.6.25) and (4.6.26) and then using the inequalities (4.6.20)
and (4.6.21), we find
I(Ynx)'(t) I -:; KQ.(t)

-:; 2AQ3(t)
OOlp (s)l l /(2-(3)dS]
2-(3
(1 Q*(S)dS)(3-l +AQ3(t)
00
Q.(t).
A similar computation for tl, t2 E [T, T
[l,
+ n]
( 4.6.27)
gives
2-(3
l(YnX)'(t2)-(ynxr(tl)1 :s; K
Ip(sW/(2-(3)dS]
+AQ3(t), (4.6.28)
370
Chapter 4
and if t E [T, T + n), we have
I(Ynx)(t) - cl
< l(Ynx)(t) - (Ynx)(T + n)1 + l(Ynx)(T + n) - cl

<
<
lOO 2AQ3(S)ds
+n 2AQ3(S)ds +
roo
JT+n
AQ3(S)ds
::; 1.
(4.6.29)
We also have
(Ynx)(t) = (Y,.,x)(T + n)
for all
t ~ T + n.
(4.6.30)
Now (4.6.27) - (4.6.30) imply that Ynx E Bn.

Finally, we shall show that Yn is a continuous map. For this, let
E Bn with Ilx - zll < 8 (8 > 0). Let
x, z
a(8) =
max
U,1I
E [c - 1, c
1]
If(u) - f(v)l
lIu - "II $ "
Now, by condition (il)' the definition of a(8), and the fact that Ix'(s)1 ::; 1
and Iz'(s)l::; 1 for all s, we find for t E [T, T + n],
l(Ynx)'(t) - (Ynz)'(t) I
:;1
:;l
00
p(s)llg(x'(s - g(z'(slds +
00
Iq(s)llf(x(s)) - f(z(slds
OO'P (s)'K x'(s)-z'(S)' [lx'(s)I,B-l+ Iz'(s)I,B-l] ds + a(8)

'
OO,q (s)'ds.
Thus, by Holder's inequality, (4.6.20) and (4.6.21), it follows that
Integrating (4.6.31) and using (4.6.20), we get
(4.6.32)
From (4.6.31) and (4.6.32) it is clear that Yn is continuous. Thus, we can
apply the Schauder fixed point theorem to obtain Xn E Bn which satisfies
Ynx n = x n . Finally, the equicontinuity of the sequence {x n } on [T,oo)
completes the proof of the theorem.
The following result is an immediate consequence of Theorems 4.6.1 and

4.6.2.
371
Theorem 4.6.3.
Let 'Y > 1, b ~ 1, f3 = min{b,2}. Assume that
p(t), q(t) E C([to, 00), lR+) and p(t) E 1/(2-/3) [to, 00). Then all solutions
of
x"(t)
+ p(t)[x'(t)[b sgn x'(t) + q(t)[x(t)[1'
oscillate if and only if
sgn x(t) = 0
It": sq( s )ds = 00.
Remark 4.6.2.
1. When the damping term is linear as in (4.1.3) and (3.1.8), in principle

any oscillation result for the undamped equation yields a corresponding
result for the damped equation. However, in practice, the problem is to
find easily verifiable oscillation criteria for damped equations.
2. In the case of nonlinear damping it is difficult to dispense with the
positivity condition on the damping. The following example illustrates this
fact.
Example 4.6.1. The damped differential equation
x"(t) - 2 ; 1 [x'(t)[b sgn x'(t)

t - nt
+ -t 211nt [x(t)[/3
sgn x(t)
0,
(4.6.33)
for b > 1, f3 = min{2, b} and t ~ 2 has an unbounded nonoscillatory

solution x(t) = t. All conditions of Theorem 4.6.1(1) are satisfied except
that p(t) is eventually negative. Also, all conditions of Theorem 4.6.1(II)
are satisfied, and hence all bounded solutions of (4.6.33) are oscillatory.

1. Theorems 4.1.1 and 4.1.2 are taken from Graef et. al. [40]. Theorem
4.1.3 is an extension of a result due to Kamenev [43] and theorem I of Wong
[53]. Theorem 4.1.4 includes theorems 1 and l' of Philos [49]. Theorems
4.1.5 and 4.1.6 are borrowed from Grace and Lalli [31]. Lemmas 4.1.1 4.1.3 are based on the work of Graef and Spikes [42]. Theorems 4.1.7 and
4.1.8 include theorems 3 and 4, respectively, of Kwong and Wong [45], and
Theorem 4.1.7 generalizes theorem 2.1 of Butler [4]. Theorem 4.1.9 is taken
from Graef and Spikes [42].
2. Conditions (F 1) - (F4) and Examples 4.2.1 - 4.2.3 are due to Philos
[48]. Theorem 4.2.1 includes as a special case the results of Grace [9,12],
and theorems 1 and l' of Philos [50]. Corollary 4.2.2 and the main result
in [48] are the same. Theorem 4.2.2 extends and improves theorems 2 and
2' of Philos [50]. Theorems 4.2.3 - 4.2.7 are due to Grace [11] and Grace
and Lalli [35]. Theorem 4.2.8 generalizes a result of Wong [54]. Theorem
4.2.9 contains theorem 2 of Philos and Purnaras [51], while Theorem 4.2.10
extends the main result of Philos and Purnaras [52]. Theorem 4.2.11 is
Chapter 4
372
new whereas Theorem 4.2.12 extends the corresponding result of Wong

[55]. Theorems 4.2.13 and 4.2.14 are also new.
3. Theorem 4.3.1 and theorem 2.2 in [4] are the same whereas Theorem
4.3.2 is due to Butler [6]. Theorems 4.3.5 - 4.3.7 extend some of the results
of Kwong and Wong [46] and are related to the work of Garce [17], also see
recent contributions of Agarwal et. a.!. [2].
4. Lemmas 4.4.1 - 4.4.3 and Theorem 4.4.2 are taken from Wong and
Agarwal [56], while Theorem 4.4.3 is new. Lemma 4.4.5 is due to Agarwal
et. al. [3]. Theorems 4.4.3 - 4.4.5 are new. Theorems 4.4.6 and 4.4.7 are
borrowed from Elbert and Kusano [7], whereas Theorems 4.4.8 and 4.4.9
are due to Kusano et. al. [44].
5. Theorems 4.5.1 and 4.5.2 are taken from Agarwal and Grace [1].
Theorem 4.5.1 extends the main result of Nasr [47].
6. The results of Section 4.6 are based on Butler [5]. For several other
interesting results in this direction, we refer the reader to the work of Grace
et. al. [8-39], and Graef and Spikes [41].
7. We note that some of the oscillation criteria of this chapter have been
extended in [8-39] to more general equations of the form
(a(t)'Ij;(x(t))x'(t))'
+ p(t)x'(t) + q(t)f(x(t)) =
0,
where a(t), p(t), q(t) E C([to,oo),JR), a(t) > 0 for t 2 to, 'Ij; E C(JR,JR+)
and f E C(JR, JR). It would be interesting to obtain oscillatory results
similar to those presented in Sections 4.2 and 4.5 for equations of the type
(4.4.1).
4.8. References
2. R.P. Agarwal, S.R. Grace and D. O'Regan, Linearization of second
order superlinear oscillation theorems, to appear.
3. R.P. Agarwal, S.-H. Shieh and C.-C. Yeh, Oscillation criteria for
second order retarded differential equations, Math. Comput. Modelling
26( 4)(1997), 1-11.
4. G.J. Butler, On the oscillatory behavior of a second order nonlinear differential equation, Ann. Mat. Pura Appl. 105(1975), 73-92.
5. G.J. Butler, The oscillatory behavior of a second order nonlinear differential equation with damping, J. Math. Anal. Appl. 57(1977), 273-289.
6. G.J. Butler, Hille--Wintner type comparison theorems for second order
ordinary differential equations, Proc. Amer. Math. Soc. 76(1979), 51-59.

7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
373
A. Elbert and T. Kusano, Oscillation and nonoscillation theorems for a

class of second order quasilinear differential equations, Acta. Math. Hungar. 56(1990), 325-336.
S.R. Grace, Oscillation criteria for second order nonlinear differential
equations, Ann. Differential Equations 4(1988), 255-264.
S.R. Grace, An oscillation criterion for second order nonlinear differential
equations, Indian J. Pure Appl. Math. 20(1989), 297-306.
S.R. Grace, Oscillation theorems for second order nonlinear differential
equations with damping, Math. Nachr. 141(1989), 117-127.
S.R. Grace, Oscillation criteria for second order differential equations
with damping, J. Austral. Math. Soc. Ser. A 49(1990), 43-54.
S.R. Grace, An oscillation criterion for second order superlinear ordinary
differential equations, Utilitas Math. 37(1990), 251-258.
S.R. Grace, Oscillation theorems for nonlinear differential equations of
S.R. Grace, Oscillation theorems for damped functional differential equations, Funkcial. Ekvac. 35(1992), 261-278.
S.R. Grace, Oscillation of nonlinear differential equations of second order,
Publ. Math. Debrecen 40(1992), 143-153.
S.R. Grace, On the oscillatory behavior of solutions of second order nonlinear differential equations, Publ. Math. Debrecen 43(1993), 351-357.
S.R. Grace, Oscillation criteria of comparison type for nonlinear functional differential equations, Math. Nachr. 173(1995),177-192.
S.R. Grace and B.S. Lalli, Oscillation theorems for certain second order
205-214.
S.R. Grace and B.S. Lalli, Oscillation of solutions of damped nonlinear
second order functional differential equations, Bull. lnst. Math. Acad.
Sinica 12(1984), 5-9.
S.R. Grace and B.S. Lalli, Oscillation and convergence to zero of solutions of damped second order nonlinear differential equations, J. Math.
Anal. Appl. 102(1984), 539-548.
S.R. Grace and B.S. Lalli, Oscillation theorems for nonlinear second
order functional differential equations with damping, Bull. lnst. Math.
Acad. Sinica 13(1985), 183-192.
S.R. Grace and B.S. Lalli, Oscillatory behavior for nonlinear second
order functional differential equations with deviating arguments, Bull. lnst.
Math. Aad. Sinica 14(1986), 187-196.
S.R. Grace and B.S. Lalli, Oscillation theorems for a second order nonlinear ordinary differential equation with damping term, Comment. Math.
Univ. Carolin. 27(1986), 449-453.
S.R. Grace and B.S. Lalli, Oscillatory behavior of solutions of second order differential equations with alternating coefficients, Math. Nachr.
127(1986), 165-175.
S.R. Grace and B.S. Lalli, An oscillation criterion for certain second order strongly sublinear differential equations, J. Math. Anal. Appl.
123(1987), 584-588.
374
Chapter 4
26. S.R. Grace and B.S. Lalli, An oscillation criterion for second order
sublinear ordinary differential equations with damping term, Bull. Polish.
Acad. Sci. Math. 35(1987), 181-184.
27. S.R. Grace and B.S. Lalli, Almost oscillation and second order functional differential equations with forcing term, Boll. Un. A1at. Ita1.
B(7)1(1987), 509-522.
28. S.R. Grace and B.S. Lalli, Oscillation theorems for second order nonlinear differential equations, J. Math. Anal. Appl. 124(1987), 213-224.
29. S.R. Grace and B.S. Lalli, Oscillation theorems for second order nonlinear differential equations with deviating arguments, Internat. J. 1I1ath.
Math. Sci. 10(1987), 35-45.
30. S.R. Grace and B.S. Lalli, On the second order nonlinear oscillations,
Bull. lnst. Math. Acad. Sinica 15(1987), 297-309.
second order nonlinear differential equations, Ann. A1at. Pura App1.
151(1988), 149-159.
32. S.R. Grace and B.S. Lalli, Oscillation in second order differential equations with alternating coefficients, Period. Math. Hungar. 19(1988), 6978.
33. S.R. Grace and B.S. Lalli, Oscillation theorems for nonlinear second
order differential equations with a damping term, Comment. Math. Univ.
Carolin. 30(1989), 691-697.
34. S.R. Grace and B.S. Lalli, Integral averaging techniques for the oscillation of second order nonlinear differential equations, J. Math. Ana1. App1.
149(1990), 277-311.
35. S.R. Grace and B.S. Lalli, Oscillation theorems for second order superlinear differential equations with damping, J. Austra1. Math. Soc. Ser. A
53(1992), 156-165.
36. S.R. Grace and B.S. Lalli, Oscillation theorems for second order nonlinear functional differential equations with damping, Comput. Math. Applic.
25(1993), 107-113.
37. S.R. Grace and B.S. Lalli, Oscillation criteria for damped functional
differential equations, Dynam. Stability Systems 9(1994), 215-222.
38. S.R. Grace, B.S. Lalli and C.C. Yeh, Oscillation theorems for nonlinear second order differential equations with a nonlinear damping term,
SIAM J. Math. Anal. 15(1984), 1082-1093.
39. S.R. Grace, B.S. Lalli and C.C. Yeh, Addendum: Oscillation theorems
for nonlinear second order differential equations with a nonlinear damping
term, SIAM J. Math. Anal. 19(1988), 1252-1253.
375-390.
41. J .R. Graef and P. W. Spikes, Asymptotic behavior of solutions of a second order nonlinear differential equation, J. Differential Equations 17(1975),
461-476.
375
42. J .R. Graef and P. W. Spikes, On the oscillatory behavior of solutions of

second order nonlinear differential equations, Czech. Math. J. 36(1986),
275-284.
43. LV. Kamenev, Oscillation of solutions of second order nonlinear equations with sign variable coefficients, Differencial'nye Uravnenija 6(1970),
1718-172I.
partial differential equations, Japan J. Math. 19(1993), 131-147.
to second order nonlinear oscillation, J. Differential Equations 46(1982),
63-77.
46. M.K. Kwong and J.S.W. Wong, Linearization of second order nonlinear oscillation theorems, Trans. Amer. Math. Soc. 279(1983), 705-722.
47. A. Nasr, Sufficient conditions for the oscillation of forced superlinear second order differential equations with oscillatory potential, Proc. Amer.
Math. Soc. 126(1998), 123-125.
48. Ch.G. Philos, A second order superlinear oscillation criterion, Canad.
Math. Bull. 27(1984), 102-112.
49. Ch.G. Philos, Integral averages and second order superlinear oscillation,
Math. Nachr. 120(1985), 127-138.
50. Ch.G. Philos, Oscillation criteria for second order superlinear differential
equations, Canad. J. Math. XLI(1989), 321-340.
51. Ch.G. Philos and LK. Purnaras, On the oscillation of second order
nonlinear differential equations, Arch. Math. 59(1992), 260-27I.
52. Ch.G. Philos and I.K. Purnaras, Oscillations in superlillear differential
equations of second order, J. Math. Anal. Appl. 165(1992), I-II.
53. J.S.W. Wong, Oscillation theorems for second order nonlinear differential
equations, Bull. Inst. Math. Acad. Sinica 3(1975), 283-309.
equations, Proc. Amer. Math. Soc. 106(1989), 1069-1077.
55. J.S.W. Wong, An oscillation criterion for second order nonlinear differential equations with iterated integral averages, Differential and Integral
Equations 6(1993), 83-91.
56. P.J.Y. Wong and R.P. Agarwal, Oscillation criteria for half-linear differential equations, Advances Math. Sci. Appl. 9(1999), 649-663.
Chapter 5
Oscillation Theory for
Sublinear Differential
Equations
5.0. Introduction
In this chapter we shall present oscillation and nonoscillation criteria
for all solutions of second order nonlinear differential equations of sublinear
type with alternating coefficients. In Section 5.1, our sublinear oscillation
results involve integrals and weighted integrals of the alternating coefficients. In some results we employ integral averaging techniques. In Section
5.2, we impose some additional conditions on the sublinear term which allow us to proceed further and extend and improve several known theorems
in the literature. In fact, we make an asymptotic study which results in
new oscillation criteria. Section 5.3 provides some new linearized oscillation
results for second order sublinear differential equations. In Section 5.4, we
shall present criteria for the nonoscillation of sublinear Emden-Fowler type
equations. In Section 5.5, we compare the oscillatory behavior of certain
nonlinear equations with the related half-linear differential equations. Here
oscillation of general nonlinear differential equations is also discussed.
5.1. Sublinear Oscillation Criteria

We shall examine oscillatory behavior of second order sublinear ordinary
differential equations of the form
+ q(t)f(x(t)) =
(a(t)x'(t))' + q(t)f(x(t))
x"(t)
and the more general damped equation
(a(t)x'(t))'
(5.1.1)
0,
+ p(t)x'(t) + q(t)f(x(t))
(5.1.2)
0,
(5.1.3)
Oscillation theory for sublinear differential equations
377
where
aCt) E C([to, 00), lR+),

(ii) pet), q(t) E C([to,oo),lR),
(iii) f E C(lR, lR), f is continuously differentiable except possibly at 0,
(i)
and satisfies
f'(x) 2': 0 for
xf(x) > 0 and

and that
f is strongly
du
j+0. feu)
xi- 0
(5.1.4)
du
feu) < 00.
(5.1.5)
sub linear in the sense that
< 00 and
-0
We shall also be interested in the prototype of these equations, namely,
+ q(t)lx(t)l"Y sgn x(t) =

(a(t)x'(t))' + q(t)lx(t)I"Y sgn x(t)
xl/(t)
and
(a(t)x'(t))'
+ p(t)x'(t) + q(t)lx(t)I"Y
0,
=
(5.1.6)
(5.1.7)
sgn x(t) = 0
(5.1.8)
respectively, where 0 < , < 1 is a constant.

We are now in the position to prove the following results.
Theorem 5.1.1. Suppose there exists a function pet) E C 2 ([to,00),lR+)
such that
p(s)q(s)ds = 00
(5.1.9)
/)0
and
00
/
a(s)p(s)
ito
00.
p(u)q(u)duds
(5.1.10)
Then the following statements hold:
(it) If f3(t) = a(t)p'(t) - p(t)p(t) :S 0 and f3'(t) 2': 0 for t 2': to, then all
bounded solutions of equation (5.1.3) are oscillatory or tend monotonically
to zero as t -t 00. If in addition, condition (5.1.5) holds, then all solutions
of (5.1.3) are oscillatory.
(h) If foo 1f3'(s)lds < 00, then every bounded solution of equation (5.1.3)
is oscillatory.
x(t) > 0 for t 2': to. Define wet) = p(t)a(t)x'(t)/f(x(t)), t 2': tl for some
tl 2': to Then for t 2': t l , we obtain
Proof.
w'(t)
<
x'(t)
-p(t)q(t)
+ f3(t) f(x(t))
-p(t)q(t)
+ f3(t) f(x(t))
x'(t)
- a(t)p(t) w (t)f (x(t))
Chapter 5
378
Integrating this inequality from tl to t, we get
w(t) ::::; w(t 1 )
it
x'(s)
t
i t1 p(s)q(s)ds + t1 (3(s) f(x(s ds.
We shall show that both conditions (id and (i2) imply that Ittl (3(s)x'(s)
/f(x(s)ds is bounded above. Since,
tl
x'(s)
(3(s) f(x(s ds
ret) du
r(t 1) du
(3(t) Jo
f(u) - (3(td J o
f(u)
-I:
(3'(s)
(foX(S)
f~:) ds
(5.1.11)
if (i 1 ) holds, then (5.1.11) yields
tl
x'(s)
r(tl) du
(3(s) f(x(s ::::; - (3(td J o
f(u) = kl'
(5.1.12)
where kl is some constant. Also, if (i2) holds, then 1(3(t)l::::; b for some
constant b> 0 and since x(t) is bounded, we find that Iox(t) du/ f(u)
is also bounded. This together with the fact that (3'(t) E C[to, 00) shows
that
t
x'(s)
t1 (3(s) f(x(s ds ::::; k2
Let k
= w(td + max{k 1 , k 2 }.
for some constant
k2 > O.
Then, we have
w(t) ::::; k - i t p(s)q(s)ds.
(5.1.13)
t1
By condition (5.1.9) there exists a sufficiently large T 2 tl

Ittl p(s)q(s)ds 2 2k for all t 2 T. Thus, it follows that
x'(t)
a(t)p(t) f(x(t
or
x'(t)
f(x(t
::::; -
::::; 1
1 it
2 a(t)p(t)
Integrating this inequality from T to t
t1 p(s)q(s)ds,
it
p(s)q(s)ds.
T), we get
S
ret) du
(x(T) du
(t 1
Jo
f(u)::::; J o
f(u) - JT a(s)p(s) t/(u)q(u)duds -+
such that
-00
as t -+
00,
379
Finally, we note that the proof of the criterion for oscillation of all
bounded solutions of (5.1.3) in the case (i 1 ) is obvious.
Theorem 5.1.2. In Theorem 5.1.1 condition (5.1.10) can be replaced by

1
00
a(s)p(s) ds
(5.1.14)
00.

x(t) > 0 for t 2: to. Proceeding as in Theorem 5.1.1, we obtain (5.1.13).
Now by condition (5.1.9), we find that w(t) -t -00 as t -t 00. Thus, there
exist a constant c> 0 and T 2: tl such that a(t)p(t)x/(t)/ f(x(t)) ::::; -c
for t 2: T, or
Proof.
x' (t)
--- < -
for
f(x(t)) -
a(t)p(t)
Integrating this inequality from t to T, we get
t 2: T.
ret) du
reT) du
rt
1
io
f(u)::::; io
f(u) - ciT a(s)p(s) ds -t
00
as t -t
00,

Corollary 5.1.1. If 0 < "I < 1 and
/00 s"Yq(s)ds
(5.1.15)
00,

Theorem 5.1.3.
replaced by
If conditions (5.1.9) and (5.1.10) in Theorem 5.1.1 are
(5.1.16)
for every constant M, then the conclusion of Theorem 5.1.1 holds.
x(t) > 0 for t 2: to. Proceeding as in Theorem 5.1.1, we obtain (5.1.13),
Proof.
i.e.,
f~~W))
::::; a(t)lp(t) [k
-1:
p(u)q(U)dU] .
Integrating this inequality from iI to t, we get
t,
x' (s)
f(x(s))ds::::;
it
t,
[it
a(s)p(s) k -
t,
p(u)q(u)du ds.
Chapter 5
380
Now from condition (5.1.16), we have
G(t)
x'(s)
f(x(s ds -+ - 00
t1
as
-+ 00,
I:S;)
G(t) =
du/ f(u). If x(t) ~ X(tl) for all large t, then
G(t) > 0 which is a contradiction. Hence, for all large t, x(t) S x(t 1), so
where
G(t) = -
(t ll
x(t)
which is again a contradiction.

Example 5.1.1.
r(t d f(u)
du
du
f(u) >
- io
> - 00,
(t 2 x'(t,+p(t)x'(t)+((1/2)+cost)lx(t)11/ 3 sgn x(t)
0, t > 0 (5.1.17)
where p(t) E C1([to,00),IRo) and p'(t) S 0 for t ~ to. It is easy to

verify all the hypotheses of Theorem 5.1.3 with p(t) = 1, and hence all
solutions of (5.1.17) are oscillatory.
Next, we present the following results.
Theorem 5.1.4. Let condition (5.1.5) hold and there exist two functions
p(t) E C 2 ([to,00),IR+) and a(t) E CI([to,oo),IR+) such that (3(t) =
a(t)p'(t) - p(t)p(t) SO, (3'(t) ~ 0, (a(t)p(t)a(t' ~ 0 for t ~ to, and
hmsup a(t)p(t)
t-4oo
If
lim
t-4oo
r a(s)ds
J~
it 1"
a(s)
t0
a(t)
< 00.
Ito a( s )ds
(5.1.18)
p(u)q(u)duds
00,
(5.1.19)

t?: to. Proceeding as in Theorem 5.1.1, we find that
inequality (5.1.13) holds for all t ~ tl' Multiplying both sides of (5.1.13)
by a(t) and integrating from tl to t, we obtain
Proof.
x(t) > 0 for
it
t1
x'(s)
a(s)p(s)a(s) f(x(s ds
it i"
t1
a(s)
t1
p(u)q(u)duds ~ k
it
t1
a(s)ds.
(5.1.20)
381
Integrating the first integral in (5.1.20) by parts, we have
ret)
a(t)p(t)a(t) Jo
du
f(u) -
1t 1
1 +
s
a(s)
t,
t,
~ k
t,
a(s)ds
1t
t,
res)
(a(s)p(s)a(s))' Jo
du
f(u) ds
(5.1.21)
p(u)q(u)duds
r:(t,)
a(t1)p(tl)a(t1) Jo
du
f(u)
Set G(t) = f;(t) du/ f(u) and n(t) = a(t)p(t)a(t). We consider two cases:
either there exists a sequence {Tn}~=l' limn->oo Tn = 00 such that
or there exists t2 2: tl such that
n(t)G(t)
-it
t,
n'(s)G(s)ds
for
t 2: t2.
(5.1.22)
In the former case, a contradiction is easily achieved upon dividing (5.1.21)

by ftt; a( s )ds and applying the hypothesis of the theorem. Otherwise,
Itt,
write C(t) =
n'(s)G(s)ds which is nonnegative and nondecreasing.
Let t3 2: t2 be chosen so large that C(t) > 0 for t 2: t3. From (5.1.22),
we deduce that n(t)C'(t) ~ n'(t)C(t), or (n(t)/C(t))' 2: 0 for t 2: t3.
Thus,
n(t)G(t)
C(t)
It follows that G(t) ~ C(t3)/n(t3)

we have
1 1
t
t,
a(s)
t,
p(u)q(u)duds
kit
Now, we can divide (5.1.23) by

obtain the desired contradiction.
t,
C(t3) ~(~t;)
for
t 2: iJ.
= kl'
where kl is a constant. Thus,
a(s)ds
+ n(t)G(tl) + kln(t).
Itt, a(s)ds
(5.1.23)
and apply condition (5.1.19) to
(tx'(t))'
+ x'(t) + ((1/2) + sint)lx(t)l""Y
sgn x(t) = 0,
t>0
(5.1.24)
where 0 < I < 1. It is easy to verify all conditions of Theorem 5.1.4 with
p(t) = l/t and a(t) = t, and hence equation (5.1.24) is oscillatory.
Chapter 5
382
In a similar way we can prove the following result.

Theorem 5.1.5. Let condition (5.1.5) hold and there exist two functions
pet) E C 2 ([to,00),lR+) and O'(t) E CI([to,oo),lR+) such that /3(t) =
a(t)p'(t) - p(t)p(t) ~ 0, /3'(t) ?: 0 and O"(t)?: 0 for t?: to, and
lim sup
t--+oo
If
t~n~
t ~
ho a(s)p(s) ds
It
to
f,tto
O'(t)
~d
a(s)p(s)
O'(s)
asp s
( ) ()
< 00.
S
to
00,
p(u)q(u)duds

x(t) > 0 for t?: to. Proceeding as in Theorem 5.1.1, we obtain (5.1.13)
Proof.
which takes the form
x'(t)
f(x(t))
+ a(t)p(t)
it p( )q( )d
t,
a(t)p(t) '
t?: tl
Multiplying both sides of this inequality by O'(t) and integrating from h

to t, we get
ret) feu)
du
it, (s) Jo(Xes) feu)
du
it a(s)p(s)
O'(S) is
ds +
p(u)q(u)duds
ret,) du
i O'(S)
O'(t) J o
t, 0'
t,
t,
a(s)p(s) ds
+ O'(tl) Jo
t,
feu)'
The rest of the proof is similar to that of Theorem 5.104.
Next we shall prove the following theorem.

Theorem 5.1.6. Suppose Q(t) =
for t?: to. If
00
ft q(s)ds exists and satisfies Q(t) ?: 0

OO
s"Y-IQ(s)ds = 00,
(5.1.25)
to

x(t) > 0 for t?: to> O. Put yet) = x(t)/t, t?: to. Then, yet) satisfies
Proof.
the equation
(t 2y'(t))' + tl+"Yq(t)y"Y(t)
O.
(5.1.26)
Set wet) = (t 2Y'(t))/y"Y(t), t?: to. Then, we have
w'(t)
tl+"Y
_ q(t) _,tl-"Y (Y'(t))2

yl+a(t)'
(5.1.27)
383
Integrating equation (5.1.27) from t to T where to::; t ::; T, we obtain
w(T) _ w(t)
T1+,
t1+,
iT(l + "(
iT iT
)w(s) d
s2+, s
Let z(t)
q(s)ds -
"(
SI-'(Y'(s))2
1+ ()
ttY's
= x'(t)/xOt(t), t
z'(t)
(5.1.28)
ds.
to. Then, we find

-
q(t) - ,,(X,-I(t)Z2(t).
(5.1.29)
Integrating equation (5.1.29) between t and T, to ::; t ::; T, we get
z(T) = z(t)
-iT -iT
q(s)ds
,,(X,-I(S)z2(s}ds.
(5.1.30)
Since ftoo q(s)ds ~ 0, we find by letting T ---* 00 in (5.1.30) that

limT-4oo z(T) = , where - 00 ::; < 00. Now, if < 0, then
z(t) < /2 for t ~ tl ~ to, and hence
t,
z(s)ds ::; -(t - tl),

2
which is impossible since x(t) > 0 for all t

lim
x(t)
CI,
t-4oo tl/(l-,)
y(t)
lim t,/(l-,)
CI,
t-4oo
where
lim w(T) =
C2
x'(t)
00.
Cl
- -,
lim t,/(I-,)
t-4OO
1-,,(
y'(t)
cn
lim t(2,-1)/(1-,)
t-4oo
1-"(
= [(1 - "()fj1/Cl-,). Therefore, we have
CI
T-4OO
where
to. Hence, 0 < f <
lim
T-4OO
T 2Y'(T) = C2,
T1+,y'(T)
("(/(1- "())c~-' ~ O. Letting T ---*
But, since
T1+,
oo w(s)
2+"ds
s
'
C3(t),
00 ::;
00
(5.1.31)
in (5.1.28), we find
C3(t) <
00.
Chapter 5
384
where
'ljJ(t) =
T w(s)
to
- 2 ds
S
yl-,(t)
1-,
1-,
it follows that C3(t) cannot be -
is bounded below on [to, CXJ),

hence
CXJ
< C3(t) <
Using (5.1.31) and (5.1.32) and taking the limit

obtain
1t
yl-,(tO)
- =--~---'-
as
CXJ,
and
CXJ.
(5.1.32)
CXJ
in (5.1.28), we
T --+
w(s)
w(t)
( 1 +s2+,
, ) - d st - < -Q(t) ,
1+, -
and hence
+ ,) w(s) ds > t,-lQ(t)
w(t) _ _ 1_1= (1
t2
t 1- , t
s2+,-
it 1=
t
to
w(s) --ds
S2
to
-1S1-,
w(u)
(1 +,)--duds?
u 2+,
t? to.
for
it
to
(5.1.33)
s,-lQ(s)ds,
i.e.,
where
C4
lit
is some constant.
-,
to
t'l=
w(s)
ds-S2
'Y
w(s)
(1+,)-ds
--+
S2+,
CXJ
as
t --+
CXJ.
(5.1.34)
Now, from (5.1.31) for any f> 0 there is T(f)? to such that
~~) ds
iT() s '
Hence, we find
'ljJ(T)
lT
I
I
w(s)d
to
T ()W(S)d
to
<
T ()
to
<
I
+ 'iT
T (<)
to
for all
w(s)d
S
S+
T? T(f).
iT w(s)
'd
2+,S S
s
iT ,s,-118 -w(u)d
- ud
2
T)
w(s)d
T
-- S s2
T(e) s2+,
T(e)
w(s)
-2-ds + 2dY for T? T(f).
-- s
T(e)
u +,
385
Therefore, we have
lim 1f;(T)
t--+oo
o.
T'Y
(5.1.35)
Now, since
using (5.1.35) and letting
T -l- 00
in (5.1.36), we obtain
(5.1.37)
Clearly, (5.1.34) and (5.1.37) give
'Ij;(t) - (1
+ 'Y)t'Y
1 1~~)'Y
00
ds
Now, since 'Ij;(t) is bounded below by
as
-l- 00
(5.1.38)
-l- 00.
- L, say, we find
Therefore, limt--+oo 'Ij;(t) = 00. Define t m , m = 1,2, by tm = sup{t :

'Ij;(t) ::; m, for m large enough}. Choose tm -l- 00 such that 'Ij;(t) 2:
'Ij;(t m ) > 0 for all t 2: t m . Then, we have
< 'Ij;(tm) - (1 + 'Y)'Ij;(tm)t~ 1.~
s~:'Y
< --'Ij;(t m ) < 0,

'Y
The following result extends Corollary 5.1.1.

Theorem 5.1.7. Assume that there exists a function p(t) E C 2 ([to, (0),
1R+) such that p'(t) 2: 0 and p"(t)::; 0 for t 2: to. If
it
lim
p'Y(s)q(s)ds
t--+oo to
00,
(5.1.39)
Chapter 5
386

x(t) > 0 for t;::: to ;::: O. Define
Proof.
w(t) =
(X(t))'
p(t)
t;::: to.
for
Then, w(t) > 0 for t;::: to. Let 0: = 1h > 1,

for t;::: to. By differentiation it follows that
~(
) (p(t)w<>(t)"
w t
then x(t) = p(t)w"(t)
(p(t)WO-l(t))"
( _0:_)
0:-1
+_l_ p"(t)w<>-l(t),
(5.1.40)
+ o:p(t)W<>-3(t)(W'(t2
t;::: to.
1-0:
(5.1.41)
Now from equation (5.1.6) and (5.1.40), we have

1
-(-)x"(t) = -(-)(p(t)wct(t)" = - p'(t)q(t)

wt
wt
for
t;::: to.
(5.1.42)
Since p"(t) SO for t;::: to and 0: > 1, the last two terms in (5.1.41) are
nonnegative, hence we may combine (5.1.41) and (5.1.42), to obtain
_0:_(p(t)Wo-1(t))" S
0:-1
- p(t)q(t)
t;::: to.
for
(5.1.43)
Integrating (5.1.43) twice from to to t, we get
p(t)w<>-l(t) ::;
Cl
+ cot -
-1) it is p'(u)q(u)duds,
0: 0:
to
(5.1.44)
to
where Co and Cl are appropriate integration constants. Clearly, (5.1.39)

implies that the right-hand side of (5.1.44) becomes eventually negative,
which contradicts the assumption that p(i)W<>-l (t) > 0 for t;::: to. This
Remark 5.1.1. In the above proof of Theorem 5.l.7 condition (5.1.39)

can be weakened to
lit is
lim sup t-'tOX)
to
to
p'(u)q(u)duds
00,
(5.1.45)
which is, in fact, sufficient to produce the desired contradiction in (5.1.44).

Example 5.1.3. Consider equation (5.1.6) with 0 < , < 1 and q(t) =
or t\ cos t, ). E lR. Condition (5.1.45) with p( t) = t shows that
(5.1.6) is oscillatory if 1 - , < ). < 1.
e' sin t
387
Oscillation theory for sublinear diHerential equations
5.2. Further Sublinear Oscillation Criteria

We assume that the function f satisfies condition (F), i.e.,
(F) f E C(1R., 1R), xf(x) > 0 and f'(x) > 0 for x
of sublinearity, namely, (5.1.5) holds.
oF 0, and the condition
Next, we let the function F(x) be defined by
du
F(x) = J+o f(u)
for
x> 0,
F(x) = [ :
f~:)
for
x <0
and (3 stands for the nonnegative constant

. { infx>o F(x)f'(x)
mm 1 + minx>o F(x)f'(x) '
-s: < 1 and F(x) > 0 for x oF O.

=lxi' sgn
x, 0 <, < 1 it follows that
=,.
Clearly, 0
f(x)
infx<o F(x)f'(x) }
1 + minx<o F(x)f'(x) .
(3
For the special case
(3
The following examples produce lower bounds for (3 which are essential
for our oscillation theory (see for example Corollary 5.2.6).
Example 5.2.1. Let f(x) = (lxi' sgn x)/(1 + Ixl'), x E 1R. We observe
that f(x) is continuous 011 1R and has the sign property xf(x) > 0
for x oF o. Moreover, f'(x) = blxl,-1 )/([1 + Ixl'12) > 0 for x oF 0 and
condition (5.1.5) holds. Furthermore, we have
F(x) = Ixl
IxI I -,
-,
+ -1-
for
x oF 0
and consequently,
inf
(3 =
(_1_ + x,)
'
x>0(I+x,)21-,
O.
Example 5.2.2. Consider the continuous function f(x) = x+lxl' sgn x,

x E 1R. Clearly, xf(x) > 0 for x oF o. Moreover, f is differentiable on
1R - {O} with f'(x) = ,Ixl,-I + 1 > 0 for all x oF O. In addition, it is
easy to verify that f satisfies (5.1.5). Next, for every x oF 0, we have
F(x)
fix i ~ <
J+0
U'
+u
fix i du =
J+0
U'
IxI 1-,
1- ,
and consequently,
inf F(x)f'(x)
x<o
1-,
inf F(x)f'(x) < inf _x_
x>o
x>o 1 - ,
[1 +,X,-I]
,
1-,
388
Chapter 5
=J 0,
On the other hand, for every x

{Ixl
F(x) = 1+0
and
F(x) -
ixi
du
1/,'Y + 'U
+0
and therefore for x
du
10
/1 U'Y
du
,+0
=J 0,
IxI1-'Y
(Ixl du
u'Y + 11, ?::
>
we find
+ 1/,
?::
(I du
Ixl::; 1
if
2u'Y = 2(1 -')')
1
2(1-')')'
1+02u'Y
if
Ixl?::
we obtain
and
F(x)1'(x)
for
Thus, F(x)f'(x) > (')'/(2(1 -')')) for all x

inf F (.' )l' (x)
x<o
=J 0,
Ixl?:: 1.
and consequently
= inf F (x) l' (x)?::

x>o
(')'
2 1 - ')'
).
Finally, we have
')'
Example 5.2.3. Consider the continuous function
Ixl'Y [k + sin(ln[l + IxID] sgn x, x E lR.

xf(x) > 0, f'(x) >
for x =J 0, and condition (5.1.5)
k?:: 1 + (111'). Moreover, for x =J 0,
f(x)
Clearly,
provided
f (x)
I
')'lxl'Y- 1
1 Ixl cos(ln[l + IxlD }

+ sin(ln[l + IxlD + ---1-1
')' 1 + x
and consequently, for x
=J 0,
Furthermore, for any x
=J 0,
F(x) =
lXI
+0
u'Y {k +
holds
du
sin(ln[l
+ u])}
>
k+
ixi du
1 +0 u'Y
1 IxI1-'Y
-----
k+11-,),
389

and
fixi
du
l+o u'Y{k + sin(ln[l
F(x) -
+ ul)}
< _1_ fix I du =

1 IxI1-'Y
k - 1 1+0 u'Y
k - 1 1- , .
Thus for all x # 0,
,[k-(l+~)]
(1 -,)(k + 1)
< F(x)j'(x) <
,[k+(l+~)]
(1 -,)(k - 1) ,
which gives
k -1 ,k - , - I
k + 1 k + 2, .
Now, we shall prove the following results.
Theorem 5.2.1. Suppose condition (F) holds, and let p(t)
1R.+ ) satisfy
[
(3p (t) +
C 2([to, 00),
(pf3(t))' ( a(t))
p(t)]2
a(t)
pf3-1(t) - p(t) a(t)
4{32
:S 1 _ {3 [(1 - (3)(P'(t))2 - p(t)p"(t)]
If
limsup
t--+oo
~
t
(5.2.1)
for
r (pf3((u))) q(u)duds =
ltD ltD
a u
t;::: to
(5.2.2)
00,

0 for t;::: to. Define w(t) = pf3(t)F(x(t)) for t;::: to. Then for
every t;::: to, we find
Proof.
x(t)
w/(t)
(3pf3- 1(t)p/(t)F(x(t))
+ (pf3(t)) (a(t)x/(t))
a(t)
f(x(t))
and
w"(t)
(3({3 - 1)pf3-2(t)(p'(t)? F(x(t))
+ (3~-l(t)p"(t)F(x(t))
+(3~-l(t)p'(t) (f~S;))) + (p:(~?)' (aj~~~~~~))

pf3 (t) [
x' (t)
I
+ a(t) -q(t) - p(t) f(x(t)) - a(t)f (x(t))
pf3(t)
f3
I
- a(t) q(t) - p (t)f (x(t))
[(
x' (t ) )
f(x(t))
X I (t))2
{pl(t)
f(x(t))
- (3 p(t)
2]
390
Chapter 5
1
]
( p{3(t))' ( a(t)) P(t)} x'(t)
aCt)
p{3(t) - aCt) f(x(t)) f'(x(t))
+(3p{3-2(t) [p(t)p"(t) - (1- (3)(pl(t))2] F(x(t))

a{3 (t)
{3
I
[
x' (t )
1
{pi (t )
- aCt) q(t) - p (t)f (x(t)) f(x(t)) - 2f'(X(t)) (3 pet)
+ (p{3(t))'
aCt)
(~) _
p{3(t)
pf3-2(t) [{ I
+ 4f'(x(t))
(3p (t)
pet)
aCt)
}]2
(pf3(t))' ( a(t))
p(t)}2
aCt)
p{3-1(t) - pet) aCt)
+4(3 {p(t)pll(t) - (1- (3)(pl(t))2} F(X(t))f'(X(t))].

Using the definition of (3, we have f'(x(t))F(x(t)) ~ (3/(1 - (3), and
hence by condition (5.2.1), it follows that w"(t)::::: -p{3(t)q(t)/a(t) for
t ~ to. Thus for every t ~ to, we obtain
it 1
:::::
p{3u)) q(u)duds
ttotoau
8
Therefore, it follows that

limsup
t-+oo
~ltlSp{3u))duds:::::
t
to a u
to
w'(to) <
00,
Theorem 5.2.2. Condition (5.2.1) in Theorem 5.2.1 can be replaced by

[
(3p (t)
for t
(a{3(t))' aCt)
aCt)
p{3-1(t)
]2
<
14~2(3
[(1- (3) (p'(t))2 - p(t)pll(t)]
(5.2.3)
to, and
pet) 2 0,
k(t)
p{3(t)p(t)/(a(t)
is a decreasing function for
t ~ to.
(5.2.4)
Proof. Let x(t) be a nonoscillatory solution of equation (5.1.3) and
assume that x(t) > 0 for t ~ to. Put wet) = p{3(t)F(x(t)) , t ~ to. Then
as in Theorem 5.2.1, for every t ~ to, we have
w
"( )
t
p{3(t) (a(t)x'(t))'
< - -'---:'-;---'--:'7-'--
aCt)
f(x(t))
391
pf3(t)
"pf3(t)
x'(t)
a(t) q(t) < - w (t) - a(t) p(t) f(x(t
for all
t ~ to
and therefore,
1: (p:(~~)
q(s)ds S; -w'(t)-w'(to)-l: k(s)
f~~~;dS
for a.ll t
~ to
Now, by the Bonnet theorem, for a fixed t ~ to and for some ~ E [to, tJ,
we find
to
x'(s)
(-k(s f(x(sds
k(to)
since
x
(t )
a
1(~)
x'(s)
= -k(t o) ito f(x(s ds
f(u)
(t O )
xCO
if x(O
du
-- <
x(to )
io
du
f(u)
du
f( ) S; k(to)F(x(to,
u
> x(to)
if
x(~)
S;
x(to)
for x > 0; also, for x < 0,

(t )
x -a
1(~)
f(u)
Hence, for every t
~
t
it 1
s
to
to
du
-- <
~
if x(~) < x(to)
[0
x(to)
du
f(u)
if
x(~) ~ x(to)
to, we obtain
(pf3((u) q(u)duds
a u
<
where M = w'(to)+k(to)F(x(to. This contradicts condition (5.2.2).
The following corollaries are immediate consequences of Theorems 5.2.1

and 5.2.2.
Corollary 5.2.1. Suppose condition (F) holds, and let p(t) E C 2 ([to, 00),
JR+) satisfy
[2p'(t) - p(t)p(t)]2 <
4(J2f.l [(1 - (J)(p'(t2 - p(t)p"(t)],

l-tJ
~ to.
Chapter 5
392
If condition (5.2.2) holds, then equation (5.1.3) with a(t) = 1 is oscillatory.

Corollary 5.2.2. Suppose condition (F) holds, and let p(t) E C 2 ([to, (0),
1R+) satisfy
p(t) ~ 0, p'(t) ~ 0, p"(t) :::; 0 and
[p,6(t)p(t)]' :::; 0 for
If condition (5.2.2) holds, then equation (5.1.3) with a(t)

Example 5.2.4.
(vtx'(t)'
Since f(x)
for every t
2yt
,
[ f3p (t)
Ixll/2sgn x, we have f3
to > 0, we find
= 1/2.
0,
Next, we let p(t)
~ to> O.
= t.
(5.2.5)
Then
(p,6(t))' a(t)
p(t)
]2
a(t)
p,6-1(t) - a(t)p(t)
~(1- sint)2
= 1 is oscillatory.
+ sin~x'(t) + (t 5 / 4 sint)lx(t)ll/2sgn x(t)
t ~ to.
4fPf3[(I- (3)(p'(t)2 - p(t)p"(t)].

1-
:::; 1 =
Also, it follows that
-I1t
t to
1
8
to
p.6(u)
(-) q(u)duds = -1 lt 1 u 5 / 4 sin ududs -+
a u t to to
8
as
00
t -+
00.
Thus, all conditions of Theorem 5.2.1 are satisfied and hence (5.2.5) is
oscillatory.
(t 2x'(t'
+ _t_x'(t) + (t ll / 4 sint)lx(t)ll/2sgn x(t) =
v1nt
As in Example 5.2.4, we have f3

all t ~ to > 1 that
1/2 and letting p(t)
t ~ to> 1.
0,
= t,
(5.2.6)
we find for
and the function p.6(t)p(t)/a(t) = 1/.jtlnt ~ 0 is decreasing. Further,
~
t
lt 1
8
to
to
(p.6((u))) q(u)duds = ~
aut
lt 1
8
to
to
U5/ 4
sin ududs -+
00
as t -+
00.
393
Therefore, all conditions of Theorem 5.2.2 are satisfied and hence (5.2.6) is
oscillatory.
We note that Theorem 5.2.2 is not applicable to (5.2.5) since condition
(5.2.4) does not hold. Also, Theorem 5.2.1 fails to apply to (5.2.6) since
condition (5.2.1) is not satisfied. We also observe that conditions of the
type Joo ds / a( s) = 00 and Joo ds / a( s) < 00 are not required.
Remark 5.2.1. Condition (5.2.3) can be simplified if a E C1([to, (0), 1R+).

In fact, it takes the form
( a'(t)) (p'(t))
a(t)
p(t)
> _(3_ (pl/(t)) + ~ (a'(t))2 for t 2': to. (5.2.7)

-
1 - (3
4(3
p(t)
a(t)
Next, we present an extension of Theorem 5.2.1.
Theorem 5.2.3. Suppose condition (F) holds, and let p(t) E C 2 ([to, (0),
1R+), satisfy (5.2.1). If
lim sup
t-+oo
t (t - s)n (p!3((s)))
q(s )ds
a s
.fto
00,
(5.2.8)
where n 2': 1 is an integer, then equation (5.1.3) is oscillatory.

x(t) > 0 for t 2': to. Define w(t) = p!3(t)F(x(t)), t 2': to. Then as
in Theorem 5.2.1 for t 2': to, we have p!3(t)q(t)/a(t) :::; -wl/(t), and
consequently for t 2': to,
l:(t-st
(P:(~~))q(S)dS
:::;
-l:(t-s)n w l/(s)dS
(t - to)nw'(to) - nit (t - s)n-1w'(s)ds

to
(t - to)w'(to) - w(t) + w(to) if n = 1
(t - to)nw'(to) - n(n - 1)
+ n(t - to)n-1w(to)
1:
(t - s)n- 2w(s)ds
if n> 1
< (t - to)nw'(to ) + n(t - to)n-1w(to).

1 it
(pf3(S))
tn to (t-s)n a(s) q(s)ds
<
(to)n,
(to)n-l w(to)
1- t
w (to)+n 1- t
-t-
394
Chapter 5
for t::2: to, and hence

limsup
t->oo
~ ltot (t -
s)n (tfi((S))) q(s)ds < w'(t o) < =,

a s
Corollary 5.2.3. Let n::2: 1 be an integer and ,\ E [0,,8]. Then equation

(5.1.1) is oscillatory if
00.
Proof. It suffices to apply Theorem 5.2.3 with
p(t)
for
e'/i3
t::2: to if f3 = 0
for t::2: to if f3 > O.
Our next result provides sufficient conditions for the oscillaion of (5.1.3)
when condition (5.2.2) fails. For this, we need to define the following functions
g(t)
p"(t)
p(t)
(1-4f32(3) (aa(t)(t))2 _ (1-f3 (3) a'(t)

p'(t)
a(t) p(t) ,
l
and
h(t)
p'(t)
p(t) -
(1(3) a'(t)
~ a(t) ,
t::2: to (5.2.9)
(5.2.10)
t::2: to
Theorem 5.2.4. Suppose condition (F) holds, and let p(t) E C 2 ([to, =),
1R+), D(t) E C([t o,oo),1R), a(t) E C 1 ([to,=),1R+) and p(t) E C([to,=),
1Ro) so that
1
lim sup -T
T->oo
rt
T
lt lt
tfi((u)) q(u)duds ::2: D(t)

au
for every
t::2: to.
(5.2.11)
Then equation (5.1.3) is oscillatory if

(5.2.12)
where D+(t)
= max{D(t), O},
t::2: to, and
= S (P((S)))2 ds <
a s
=,
(5.2.13)
395

limsupth(t) <
h2(t):S -cg(t) and h2(t)+h'(t) ~ get) for t ~ to
00,
t-HX)
(5.2.14)
for some positive constant c.

Let x(t) be a nonoscillatory solution of equation (5.1.3), say
for all t ~ to. Define wet) = p!3(t)F(x(t)) for t ~ to. Then for
every t ~ to, we have
Proof.
x(t)
01 0
_
w'(t) -
p'(t)
x'(t)
!3
13 pet) wet) + p (t) f(x(t))
and
p!3(t) (a(t)x'(t))' _ !3(t)f'(x(t))

aCt) f(x(t))
p
w"(t)
+(
p!3(t))'
x'(t)
aCt)
aCt) f(x(t))
(~)2
f(x(t))
p'(t) ,
+ 13 pet) w (t) + 13
(P'(t))'
pet) wet)
p!3(t)
!3 pet) x'(t)
- aCt) q(t) - p (t) aCt) f(x(t))
_f'(X(t))F(X(t))w~t)
[Wl(t)
-f3~g?W(t)r
+ [f3 PI (t) _ a/(t)] [wl(t) _ f3 PI (t)w(t)]

pet)
aCt)
pet)
+f3p l(t) w'(t) + 13 [p"(t) - (pl(t))2] wet).

pet)
pet)
pet)
Using the definition of the number 13 and applying the method of completing the square, we obtain for t ~ to,
w"(t) < _p!3(t) q(t) _

aCt)
+
-
~(t) (p(t)) ~
aCt)
f(x(t))
(113 ) 1 [,
{ (t) (1 - 13) (t) } ]
13 wet) w (t) - pet) - 2i3 aCt) wet)
13 [pll(t)
pet)
13) (a /(t))2 _ (~) (a/(t)) Pl(t)] wet)

+ 4132
aCt)
13
aCt) pet)
pi
( 1-
a'
or
w" (t)
<
ret)
- aCt) q(t) -
~(t) aCt)
pet) x'(t)
f(x(t))
+f3g(t)w(t) -
13 13 ~() [w'(t) - h(t)w(tW

1- wt
(5.2.15)
Chapter 5
396
- liT,
Integrating (5.2.15) twice and then multiplying it by

for T::::: t ::::: to,
t) w (t)
we obtain
iT is --q(u)duds
pf3(u)
a(u)
x'(u)
1 iT is
+ T1 iT is ~(u) (p(u))
a(u) f(x(u)) duds + {3T
-g(u)w(u)duds
-weT)
- - + -wet) + ( 1 T
+(
(3
) 1
1 - (3
IT IS
t
::::: -1
T
1
w(u)
[w'(u) - h(u)w(uWduds.
(5.2.16)
Now, taking into account condition (5.2.13) and using the Bonnet theorem,
we conclude that for any s, t with s::::: t ::::: to, there exists a number
~ E [t, s] such that
x' (u) )
is (_13 (u) p(a(u)u) ) f(x(u))
x' (u)
i s(_13 (u) p(a(u)u)) ( f(x(u))
du =
du
_13 pet) 1~ f(x(u))
x'(u)
_13 (t) pet)
1t x'(u)
= - (t) aCt)
du =
aCt) ~ f(x(u)) du
t
fJ
fJ
= ~(t)p(t)
fJ
(t)
3:JL :::;
pl3(t)p(t) F(x(t))
aCt)
aCt) x(~) fey)
since
-fJ
(t)
fey) <
x()
if x(~) > x(t)
dy
Pa((tt))w(t),
io
x(t)
dy
fey)
if
x(~):::; x(t)
for x> 0, and
(t)
dy
x(~) fey)
for x < O. Hence, for
_ weT)
T
>
+ wet) +
T
if x(~) < x(t)
< {
10
x(t)
T::::: t ::::: to,
dy
fey)
if
x(~)::::: x(t)
inequality (5.2.16) becomes
(1-.!.)
[Wl(t) + pet) wet)]
T
aCt)
~ iT is ~(~~) q(u)duds + {3~ iT is (-g(u))w(u)duds

+ ( 1-(3 (3 ) T1
iT is w(u)
1
[w'(u) - h(u)W(U)]2duds.
t
(5.2.17)
397
Next, by (5.2.11) and (5.2.14) inequality (5.2.17) gives
w'(t)
:~!~ w(t)
> D.(t) + l'r:./{J/ w~) +
{3 1
+-(3
1-
00
{31O (-g(u))w(u)du
1 [w'(u) - h(u)W(u)]2du
-(-)
w u
for all t 2: to, which proves that
+ :~!~ w(t) 2:
w'(t)
D.(t)
for every
t 2: to,
. . w(T)
hl1l1I1f - T < 00,
1.
and
1.
(5.2.19)
T-+oo
00
(5.2.18)
(-g(u))w(u)du < 00,
(5.2.20)
to
00
to
1
-(-) [w'(u) - h(u)w(uWdu
W tL
< 00.
(5.2.21 )
By (5.2.14), for every T 2: to, we find
1.
to
_(1) [w'(u) - h(u)w(uWdu

w u
1.
1.
T [w'(u)J2
iT
iT
( ) du - 2
h(u)w'(u)du +
h2(u)w(u)du
~
wu
~
~
T [w'(uW
( ) du - 2h(T)w(T) + 2h(to)w(to)
w u
to
1:
+
>
(T[h2(u)
+ 2h'(u)Jw(u)du
ito
[w~~~J2 du -
- (c + 2)
1.
2Th(T)
(w~)) + 2h(to)w(to)
(-g(u))w(u)du.
to
Therefore, it follows that
) [w'(u) - h(u)W(u)]2du
i toT _(1
w u
2:
1.
00
to
[W'~Ur
du - 2 [lil1lS11 P Th(T)]
w u
T-+oo
+ 2h(to)w(to) -
[lil1linf WT(T)]
T-+oo
(c + 2) {oo (-g(u))w(u)du,
ito
Chapter 5
398
which because of (5.2.14) and (5.2.19) - (5.2.21) yields
00
. to
[u,t(u)F
( ) du <
W U
Furthermore, by the Schwartz inequality, for t
w(t)
:::;
:::;
to, we have
+ {[w(tW/ 2 - [W(tO)P/2}
[[u,(toW/2
:::;
(5.2.22)
00 .
2w(to)
+ 2 [[w(t)]1/2 -
2w(to)
-+:2 . t.o [w(ulP/ 2du
2w(to)
+~
2w(to)
and consequently,
[w(to)p/2f
u,'(u)
(It. It
1
+ -t
11
2w(to)
( ) <t { wt
- + -2
to
-
[It.
00
to
dU)
t.o
00
t.o
t.o
]2
[w'(u)F du
'w(u)
[w'(ulF
( ) du
W U
[w'(u)F du }
w(u)
for
to.
(5.2.23)
Finally, by using (5.2.18), (5.2.22) and (5.2.23), we get
where
= 2w(to) + ~
to
00
t.o
[w'(s)j2 ds <
w(s)
00.
This in view of (5.2.12) and (5.2.13) leads to a contradiction.
Remark 5.2.2. When a(t) == 1 the functions g and h defined in

(5.2.9) and (5.2.10), respectively, are reduced to g(t) = p"(t)/ p(t) and
h(t) = p'(t)/ p(t) for t ~ to, and hence condition (5.2.14) becomes
(p'(t))2 :::; - cp(t)p"(t)
for some constant c> 0 and
t ~ to. (5.2.24)
399

Also, by condition (5.2.24), we observe that
p(t) = p(to)
it
p'(s)ds > (t - to)p'(t),
to
Therefore, limsupt-4oo(tp'(t))jp(t) <
~ to.
00.
The following three corollaries are immediate.

Corollary 5.2.4. Let condition (F) hold, and let p(t) E C 2([to, (0), lR+),
D(t) E C([to,oo),lR) and p(t) E C([to,oo),lRo) be such that
1
lim sup -T
T-4OO
iT 1
8
pJ3(u)q(u)duds
D(t)
for every
~ to.
(5.2.25)
Then equation (5.1.3) with a(t) = 1 is oscillatory if conditions (5.2.12)

and (5.2.24) are satisfied, and
(pJ3(t)p(t))' :::; 0 for
~ to
and
Joo sp2(s)ds <
00.
Corollary 5.2.5. Let condition (F) hold, and let p(t) E C 2 ([to, (0), lR+),
and D(t) E C([to, (0), lR) be such that condition (5.2.25) hold. Then
equation (5.1.1) is oscillatory if conditions (5.2.12) and (5.2.24) are satisfied.
0:::; >. < fJ
Corollary 5.2.6.
Let condition (F) hold, and let
DE C([to,oo),lR) be such that
liTiS
lim sup -T
T-4OO
uAq(u)duds ~ D(t)
and
t ~ to
for every
Then equation (5.1.1) is oscillatory if condition (5.2.12) holds.

Proof.
p(t)
The proof follows immediately from Theorem 5.2.4 by letting

for t ~ to.
= tA/ J3
Next, we shall prove the following oscillation result for (5.1.3).

Theorem 5.2.5. Let condition (F) hold, and let p(t) E C 2([to, (0), lR+),
D(t) E C([to,oo),lR), a(t) E CI([to,oo),lR+) and p(t) E C([to, (0), lRo)
be such that
1
liminf -T
T-4OO
iT 1
8
p/3((u)) q(u)duds
tau
D(t)
for every
~ to.
(5.2.26)
Then equation (5.1.3) is oscillatory if condition (5.2.14) holds,

lim sup
t-400
[it
to
Sh 2(S)dS]
-lit
to
[D+(S)]2 ds
S
00
(5.2.27)
400
Chapter 5
(rJ3(t)p(t)/a(t))' ~ 0 for t ~ to, also
and
limsup
t-4CXJ
[I Sh2(S)dS] -ll
t
to
to
s (P((S)))2 ds <
a s
00.
Proof.

0 for t ~ to. Define w(t) = p!3(t)F(x(t)), t ~ to. Then as in
Theorem 5.2.4, we obtain for T 2: t ~ to that
x(t)
i-
w'(t)
+ :~~~ w(t)
2:
O(t)
+ lim sup wT(T) +
+--(3(3 1 -(-)
CXJ
1-
Also,
lim sup
(31
CXJ
(-g(s))w(s)ds
T-4CXJ
1 [w'(s) - h(s)w(s)J 2ds.

w s
we::)
<
(5.2.28)
00
T-4CXJ
also (5.2.18), (5.2.20) and (5.2.21) hold. Furthermore, for T 2: to, we find
T [w'(s)JZ ds =
) [w'(s) - h(s)w(sWds + iT 2h(s)w'(s)ds
i toT ~(
w(s)
w s
to
to
_ (T h2(s)w2(s)ds
Jto
LT
w~s) [w'(s) -
h(s)w(s)J 2ds
+ 2Th(T)
-2h(to)w(to) - 21T h'(s)w(s)ds

to
< loT
-I
(w~))
h2(s)w(s)ds
to
w~S)[w'(S)-h(s)w(sWdS+2Th(T) (w~))
-2h(to)w(to) + 21T -g(s)w(s)ds + lT h2(s)w(s)ds

to
lT
to
~(
) [w'(s) w s
-2h(to)w(to)
to
h(s)w(s)fds + 2M(Th(T))
+ 21T -g(s)w(s)ds + M lT Sh2(s)ds,

to
to
where
sup w(T)
T?to
<
00
(by (5.2.28)).
(5.2.29)
Oscillation theory for sublinea.r differential equations
401
Hence, taking (5.2.20), (5.2.21) and condition (5.2.14) into account, we

conclude that there exists a positive constant N such that
[w'(sW
,( ) ds ::::: N
to
lt
to
11.. 8
Sh2(S)ds
for
t 2:: to.
(5.2.30)
Finally, from (5.2.18) and (5.2.30) for t 2:: to, we get
< Ml . ~(1) [U"(8)

t
to 1L S
+ P((S))W(8)12 d8
a s
< 2Mjt [n/isfd8+2Mlt (P((S)))2 w(s)ds

n'
to
a s
to
< 2MNlt Sh2(S)dS+2M2lts (p((s))) ds,

to
a s
to
where NI is as in (5.2.29). Thus, it follows that
[I Sh2(S)dSl-llt [!1+(sW ds
t
to
to
::::: 2MN + 2M2 [1>h 2(S) dS
r1> (~~:~r
1
ds <
==
If a(t)
Remark 5.2.3.
reduces to
p'(t) 2:: 0
00,
1 in Theorem 5.2.5, then condition (5.2.14)
and
p"(t) ::::: 0
t 2:: to.
for
(5.2.31 )
Now, we present the following immediate corollaries.

Corollary 5.2.7. Let condition (F) hold, and let p(t) E C 2 ([to, oo),lR+),
!1(t) E C([to, (0), ffi) and p(t) E C([to, 00), ffio) be such that
1
liminf -T
T-+oo
j.T IS pf3(u)q(u)duds
t
Then equation (5.1.3) with a(t)

holds,
lim sup
t-+oo
2:: !1(t)
[I
to
(p'(S))2
-(-) ds
P
for every
t 2:: to.
= 1 is oscillatory if condition (5.2.31)
j-l lt
to
[!1+(s)j2
S
ds
(5.2.33)
00
and (pf3(t)p(t))'::::: 0 for t 2:: to, also

limsup
t-+oo
[I
to
S (p'((s))) dsj-l lt sp2(s)ds <

P
(5.2.32)
to
00.
Chapter 5
402
Corollary 5.2.8. Let condition (F) hold, and let pet) E C 2 ([to, (0), lR+)
and net) E C([to, (0), lR) be such that condition (5.2.32) hold. Then
equation (5.1.1) is oscillatory if conditions (5.2.31) and (5.2.33) are satisfied.
Corollary 5.2.9. Let condition (F) hold, and let n E C([to, (0), lR) be
such that
liTiS
liminf -T
T->oo
uf3 q (u)duds ~ net)
for every
to.
(5.2.34)

.
it
hmsup t->oo In t to
[n+(s)F
S
ds =
(5.2.35)
00.
Proof. It suffices to apply Theorem 5.2.5 with pet) = t for t
to.
ReIllark 5.2.4. Conditions (5.2.31) and (5.2.35) imply (5.2.33). Indeed,

from (5.2.31) it follows that limsuPT->oo Tp'(T)/p(T) < 00 (see Remark
5.2.2), and hence there exists a positive constant b such that tp'(t)/ pet) ~
b for all t ~ to. Thus, for all large t, we have
t S
to
it
(p'(S))2 ds ~ b2
pes)
to
ds = b2 [lnt- Int ol
< 2b2lnt,
which proves the proposed assertion.

ReIllark 5.2.5.
As we have mentioned earlier, the function f(x) =
Ixl"! sgn x, x E lR (0 < 'Y < 1) is a classical example which satisfies
condition (F). Other such functions are:
(II) f(x) = Ixl"! sgn x + x, x E lR (0 < 'Y < 1) with 'Y/2 ~ /3 < I,
(h) f(x) = Ixl l / 2 sgn x/(I + Ixl l / 4 ), x E lR with /3 = 1/4,
(13) f(x) = Ixl"![k+sin(ln[I+lxl])] sgn x, x E lR (0 < 'Y < I, k ~ I+(1h))
with (k -I)("(k - 'Y -I)/(k + I)(k + 2'Y) ~ /3 < 1.
Also, see Examples 5.2.1 - 5.2.3.
ExaIllple 5.2.6. Consider the differential equation
(to:x'(t))'
+ tOx'(t) + t'>'(sint)f(x(t))
t ~ to> 0
0,
(5.2.36)
where a, (J and A are real constants, and the function f is anyone

on the above in the Remark 5.2.5. We let - a + /3 + A = 8, 0 < 8 ~ I,
(J - ex + /3 ~ 0 and (J - a < -1. Now,
get) =
a2
[ 4/3
2
1)
a - /3] (1-/3) ( t 2
and h(t)
/3 - a(I2/3
(1)
/3)] t '
403

We also let
C[0 +,6 -~]

4,62
for some constant
>0
and
For all T, t with T
1jTjS
to, we obtain
ui5 sin ududs
- T i5 - 1 sin T - 20Ti5 - 2 cos T
+~
[t i5 sin t
+ 28t"-1 cos t -
T---+oo
jT
s"-3cos sds.
to, we find
1jTjS
lim sup -T
s"-3 sin sds
[ttl cost - ot"-l sin t - 0(0 - 1)t,,-2 cost]
-0(0-1)(0-2)
Thus, for t
1)(0 + 1)Ti5 - 3 sin T
0(0 - 1)(0 + 1)t,,-2 sin t]
1jT
- 0(0 - 1)(0 + 1)(0 - 2) T
+ (1 - ~ )
+ 0(0 -
uf3 q (u)duds
1jT j"
lim sup -T
T---+oo
utI sin ududs
> ttl cos t - ot"-l sin t - 0(0 - 1)t"-2 cos t

-0(0 -1)(0 - 2)
> t i5 cost -
00
s"-3cossds
Jl,
where Jl is a positive constant. Clearly, condition (5.2.11) is satisfied with

O(t) = t i5 cos t - Jl, t ~ to. Next, we consider an integer N such that
2Nrr - (7r/4) ~ max {to, (1 + V2/L)1/i5}. Then for all integers n ~ N, we
have O(t) ~ 1/V2 for every t E [2n7r - (7r/4),2n7r + (7r/4). Thus, we
obtain
1
-2:
2
00
n=N
2mr +(1l'/4)
2mr-(7I'/4)
-1
LIn
2
00
n=N
-ds
S
2]
1+-8n-1
00,
Chapter 5
404
i.e., condition (5.2.12) is fulfilled. Hence, Theorem 5.2.4 can be applied to

guarantee the oscillation of equation (5.2.36).
x"(t) + t[t + lntJ-'(sint)lx(t)I' sgn x(t)
(5.2.37)
0,
where 0 < 'Y < 1, t 2: to > 1. Define pet) = t + In t, t 2: to and

observe that condition (5.2.31) is satisfied. Furthermore, for every T, t
with T 2: t 2: to, we have
~lT1S rJ(u)q(u)duds
- sinT + (1-
~)
~lTls usinududs
[tcost - sintJ +
~[-2coST + tsint + 2sintJ
and consequently, for every t 2: to,

1
liminfT
T--+oo
1l
T
pf3(u)q(u)duds = tcost-sint-l 2: tsint-2.
Thus, condition (5.2.34) holds with net) = t cos t - 2, t 2: to. We consider

a number tl such that tl 2: max{ to, 4V2}. Next, we choose an integer
N such that 2N7r - (7r/4) 2: tl' Then for every integer n 2: N, we have
net) 2: t/(2V2) for t E [2n7r - (7r/4),2n7r + (7r/4)J. Thus, for n 2: N,
we get
2n7r +(7r/4) [n+(s)j2
-=----'-~ds
to
>
>
1
1
2n7r +(7r/4) [n+(s)j2
2n7r-(7r /4)
S
2n7r
1
+(7r/4)
sds
8 2n7r-(7r/4)
and therefore,
.
1
hmsup t--+oo In t
it
to
1
n--+oo In [2n7r + ~]
7r 2 n
> lim
- n--+oo 8In [2n7r + ~J
[n+(s)j2.
ds > hmsup
S
ds
2n7r +(7r/4) [n+(s)j2
to
ds
00.
Hence, condition (5.2.35) is satisfied, and consequently condition (5.2.27)

holds, as noted in Remark 5.2.4. Therefore, by Theorem 5.2.5 the equation
(tx'(t))' + ~x'(t) + (t 3 / 2 sint)\x(t)\1/2 sgn x(t)

t
0,
2: to> 1. (5.2.38)
405
Define p(t) = t, t 2: to. Then, g(t) = -1/(2t 2) and h(t) = 1/(2t),

and hence condition (5.2.14) is satisfied. As in Example 5.2.7, we find that
O(t) = tcost-2, t 2: to. Thus, all conditions of Theorem 5.2.5 are satisfied,
and therefore equation (5.2.38) is oscillatory.
In what follows it will be convenient to use the following notation. For

p(t) E C 2([to, 00), IR+) and t 2: to, we let
",
(t) =
h1(t) = f'(x)F(x)
gl(t)
f3P;~~~) -
f3",(t)
a(t) (a!3(t))' p(t)

p!3(t) a(t)
- a(t)'
p' (t )
[ I I p ' (t)
2",(t)
~g; -
+ 2 13 p(t) + 13 p(t) f (x)F(x) ,
13[1 + 13]' (x)F(x)]
h () = 1-13 [~ ()
2 t
13
2'" t
and
pll (t) _ f3",(t) p'(t) _ 13

[f3 p(t)
p(t)
(~gD
+ ]'(x)F(x)hi(t),
~f3(l+f3) pl(t)]
+2
1 - 13
p(t)
[1 + L]
(p'(t))2 + _f3_ h 2(t)].
1 - 13
p(t)
1 - 13
2
We are now in the position to prove the following result.

Theorem 5.2.6. Suppose condition (F) holds with 13 > O. Let n 2: 1
be an integer and p(t) E C 2 ([to,00),IR+) satisfy
limsup th 2(t) < 00,
h~(t):S
- Cg2(t)
and
h~(t)
+ h~(t) 2: g2(t)
t~oo
(5.2.39)
for some constant c> 0 and t 2: to. Then equation (5.1.3) is oscillatory
if there exists a function 0 E C([to, 00), IR) satisfying (5.2.12) and
1
lim sup -;t~oo
Jt (t - s)n_(
p!3(s)
-)
T
a s
q(s)ds 2: O(T)
for every
T 2: to. (5.2.40)

x(t) =1= 0 for t 2: to. Moreover, let w(t) = pf3(t)F(x(t)), t 2: to. Then for
t 2: to, we have
Proof.
w'(t) = f3pl(t) w(t)

p(t)
and
x'(t)
pf3(t) f(x(t))
p!3(t) a(t)x'(t)
a(t) f(x(t))
w'(t) -
f3~g} w(t),
406
Chapter 5
w"(t) = (3pl/(t) w(t) _ {3 (p'(t)) 2 W(t)

p(t)
p(t)
+ (3P'(t) W(t) + (pfJ(t))' a(t)x'(t)

p(t)
a(t)
f(x(t))
p{3(t)
f3 p(t) x'(t)
1 [,
P'(t)] 2
- a(t) q(t) - p (t) a(t) f(x(t)) - W(t) W (t) - (3 p(t) W(t)
x f'(x(t))F(x(t))
(5.2.41)
= - pf3(t) q(t) + 1/(t) [W'(t) - (3P'(t) W(t)] + (3P"(t) w(t)

a(t)
p(t)
-{3 (p'(t))2 w(t)
p(t)
+ (3P'(t) w'(t) _ f'(x(t))F(x(t)) (W'(t))2

p(t)
w(t)
+2{3~g; f'(x(t))F(x(t))w'(t) = _pf3(t) q(t) _ _ l_[w'(t) _

a(t)
-(3[1
p(t)
w(t)
+ (3f'(x(t))F(x(t))]
(32 f'(x(t))F(x(t))
(~gD 2w(t)
hI (t)w(t}f [(3P"(t) _ (31/(t) (p'(t))
p(t)
p(t)
(~g;) 2 + f'(X(t))F(X(t))h~(t)l w(t)
pf3(t)
1
,
2
a(t) q(t) - (-91(t))W(t) - w(t) [w (t) - hI (t)w(t)]
X
f'(x(t))F(x(t)).
By the definition of (3 and (5.2.41), we find that for

the form
t:::: to,
(5.2.42)
(5.2.42) takes
w"(t) :s; - pf3t)) q(t) - (-92(t))W(t) - ~~() [w'(t) - h2(t)W(tW.

at
1-,uwt
(5.2.43)
Thus, for any t, T with t:::: T :::: to, we obtain
(t -
:s; -
s)nP:(~) q(s)ds +
{3
1-,u
--(.I
ht
ht
T
(t - s)n( -92(S))w(s)ds
1
(t - sr-(-)
[w'(s) - h2(S)W(sWds
WS
(t - s)nw"(s)ds = (t - T)nw'(T) - n
(t - T)w'(T) - w(t)
ht
+ w(T)
ht
(t - s)n-1w'(s)ds
if n = 1
(t - TY'w'(T) - n(n - 1) [(t - s)n- 2w(s)ds + n(t - T)n-lw(T)

if n> 1.
407
Therefore, for t 2: T 2: to, we have
~ t
tn
iT
(t - st p!3((s)) q(s)ds + ~
(t - s)n( -g2(s))w(s)ds
as
t n iT
iTt (t -
+ -.L~n
l-(3t
~) w'(T) - w(t) + w(T)
(1-
(1 - f)
<
sr-1-[w'(s) - h2(S)W(s)fds
w(s)
if n = 1
1) t: It (t - s)n- 2w(s)ds
w'(T) - n(n -
T)n-l w(T)
+n ( 1 - -t
t
Thus, for every T 2: to, we get
1
lim sup --;:;
t---'>oo
(3
+ --(3
1-
It
1
pf3(s)
(t - s)n_(
-) q(s)ds +
00
00
if n > l.
(-g2(s))w(s)ds
-1() [w'(s) - h2(S)W(s)fds
w'(T) -liminf w(t) if n = 1

t---'>oo
t
{
<
t
w'(T) - n(n - 1) liminf ~ { (t - s)n- 2w(s)ds
t---'>oo t n iT
if n> 1
and consequently, by condition (5.2.40), for all T 2: to, we obtain
w'(T) 2: O(T)
(3
-g2(s)w(s)ds + -{3
00
1-
liminf w(t)
t---'>oo
00
-(-) [w'(S)-h2(S)W(S))2ds
W S
if n = 2
n(n - 1) liminf -1
t---'>oo t n
lt
T
(t - s)n-2w(s)ds if n > 2.
This shows that w'(T) 2: O(T) for every T 2: to,

roo
-g2(s)w(s)ds <
00
00,
to
ito
-(-) [w'(s) - h2(S)W(sWds <

W S
00
and
liminf w(t) <
{
t---'>oo
liminf
~n t
t---'>oo
ito
00
if n
=1
(t - s)n- 2w(s)ds <
(5.2.44)
00
if n> l.
408
Chapter 5
Now, we shall show that (5.2.44) implies
. f w(t)
1Imlll
-- <
t->oo
t
(5.2.45)
00
in both cases where n = 1 or n > 1. Clearly, this is true for n =

1. So, we consider the case where n > 1. Suppose (5.2.45) fails, i.e.,
limHoow(t)/t=oo. Let a be a number such that 0<a<1/(n(n-1)),
and p, be an arbitrary positive constant. Then there exists a
to
such that w(t)/t::::: p,ja for all t::::: T I . Thus, for t::::: T I , we have
TI : : :
-1
tn
it
to
jt
jt (
at
1
-;;;
>
(t - s)n- 2 w(s)ds
(t - s)n- 2 w(s)ds
Tl
P, 1
> --;;;
T,
t-s )n -2sds
T )
1 (
; [ n-1 1--f
n-I -:n1 ( 1--f

T ) n]
.
Since,
n-I
lim [ -1- ( 1 - T
-.!. )
- -1 ( 1 - T
-.!. )
t->oo n - 1
t n t
we can choose a T 2
:::::
n(n - 1)
TI so that
_1_ (1- TI)n-l _ ~ (1- Tl)n

n-1
n]
t n t
>
for
all
> a,
t::::: T2.
Now, it follows that
-1
it
t n to
(t - sr- 2 w(s)ds > p,
which gives
lim -1
t->oo t n
since p,
follows.
it
to
for every
(t - s)n-2w(s)ds
t::::: T2 ,
00,
> 0 is arbitrary. This contradicts (5.2.44), and hence (5.2.45)
The remainder of the proof proceeds exactly as in Theorem 5.2.4 with

the functions h(t) and g(t) replaced by h1(t) and g2(t) respectively,
and hence we omit the details. This completes the proof.
For the special case when p(t) = 0 and a(t) = 1, we have
'YI(t)
'f
(3P'(t)
p(t) ,
h (t) =
2
p'(t)
p(t)
and
p"(t)
g2(t) = (3 p(t) .
(5.2.46)
409
Thus, the following corollary is immediate.

Corollary 5.2.10. Suppose condition (F) holds with (3 > O. Let n 2: 1
be an integer and the function p(t) E C 2 ([to, 00), 1R+) satisfy condition
(5.2.24). Then equation (5.1.1) is oscillatory if there exists a function n E
C( [to, 00), IR) such that conditions (5.2.12) and (5.2.40) hold.
The following result is similar to that of Theorem 5.2.3.
(Fl
Theorem 5.2.7. Suppose condition

holds. Let n 2: 1 be an integer
and the function p(t) E C 2 ([to,00),1R ) be such that 92(t) :s; 0 for
t 2: to. If condition (5.2.8) holds, then equation (5.1.3) is oscillatory.
0 for t 2: to. Moreover, let w(t) = pf3(t)F(x(t, t 2: to.
Then as in Theorem 5.2.6, (5.2.43) holds. Thus, for t 2: to, we have
w"(t) :s; -pf3(t)q(t)/a(t). The rest of the proof proceeds exactly as in
Theorem 5.2.3, and hence we omit the details.
Proof.
x(t)
i=
The following corollary of Theorem 5.2.7 is immediate.

Corollary 5.2.11.
Suppose condition (F) holds. Let n 2: 1 be an
integer and the function p(t) E C 2 ([to,00),IR+) be such that p'(t) > 0
and p"(t) :s; 0 for t 2: to. If condition (5.2.8) holds, then equation (5.1.1)
is oscillatory.
Next, we shall prove the following oscillation result for (5.1.6).
Theorem 5.2.8. Suppose that
ll l
T
lim -T
T-+oo
to
and define
Q(t)
If
l
lim -T
T-+oo
u"Yq(u)duds
[T [S u"Yq(u)duds
00
to
to
-[Q(s)
S
+ k(S)]2ds
1R
(5.2.47)
t 2: to.
(5.2.48)
exists in
for
for every function k(t) E C([to,oo),IR) with

00
(5.2.49)
limHoo k(t) = 0,
then

x(t) > 0 for t 2: To ~ to. Define w(t) = t"Yx1-"Y(t)/(1- 'Y), t 2: To. Then
for t 2: To, we have
Proof.
410
Chapter 5
and
w'(t)
'-t-
w" (t)
- 'wet)
T + ,{y-1x-'Y(t)x'(t)
+ t'Y x-, (t )x" (t)
-,t'Y x-'Y- 1 (t)( x' (t))2
+, [Wl(t)
,Wl(t) _,wet)
t
t2
_,wet)]
t2
[Wl(t) _,w(t)]2 +t'x-,(t)x"(t)

t
__
, __1_
l-,w(t)
t'x-,(t)x"(t) _ _
, __1_ [Wl(t) _ wet)] 2
l-,w(t)
t
+ _,__1_ [Wl(t) - W(t)]2
-w"(t) = t'Yq(t)
This gives
_ weT)
T
u'q(u)duds
+1
=, ~ iTis w~u)
[WI(U) _
for every T, t with T 2: t 2: To. Therefore, for every t 2:
w'(t)
t 2: To.
for all
(1 _.!.)
w/(t)
T
+ wet) +
iTis
l-,w(t)
11 1
, 1 -+-lim _T
T-too
To,
duds
we obtain
weT)
+ T-too
lim T
[,
W(S)]2
w (s) - - ds
00
1-,
u'q(u)duds
w~u)]
w(s)
and consequently,
<
lim weT)
T-too
and
00
to
_1_ [WI(S) _ W(S)]2 ds
w(s)
(5.2.50)
00
<
00.
(5.2.51)
Hence, we have
w'(t)
where
G(t)
Q(t)
lim wT(T)
T-too
_'_1
1-,
00
+ G(t)
for every
t 2: To,
_1_ [WI(S) _ W(S)] 2 ds,

w(s)
s
t 2: To.
(5.2.52)
411
Now, we set
w(T)
k(t) _
-
T--+oo
w(t)
+ G( t)
lim - T
w(T)
+ G(to)
lim - T
T--+oo
if
- --
w(to)
- --
> to
if
to
to:::; t:::; To.
In view of (5.2.50) and (5.2.51), k(t) E C([to, (0), JR) and limHoo k(t) =
O. Furthermore, from (5.2.50) there exists a positive constant. K such
that
w(T)
-T :::;
T 2: to
for all
(5.2.53)
Thus, from (5.2.52) and (5.2.53), we get
[WI(S) _ W(S)]2 ds
roo _1_
.fTo w(s)
roo _(1)
.fTo W S
00
To
[Q(s)
+ G(s) _ W(S)]2 ds
lim w(T)
T
T--+oo
1
-(-) [Q(s)
w s
+ k(s)J 2ds 2: K
00
To
-[Q(s)
S
+ k(s)fds
and therefore, from (5.2.51), we find
001
-[Q(s)
to
+ k(S)]2ds
:::;
TO
to
+K
+ k(sWds
-[Q(s)
s
r= w(s) [WI(S) _ W(S)]2

ds
s
_1_
lTo
<

Example 5.2.9.
x"(t)
00,
+ (tA sin t)lx(t)I'
sgn x(t)
0,
t 2: to > 0
(5.2.54)
where 0 < 'Y < 1 and - 'Y < .\ < 1 - 'Y. We set 5 = 'Y +.\ so that
0<5 < 1. Now, for any T, t with T 2: t 2: to, we have
T1
jT jS u'q(u)duds
t
T1
jTjS
t
- T O- 1 sin T - 25TIi - 2 cos T
Osinududs
+ 5(5 -
+ T [t li sin t + 25t1i -1 cos t T
sin T
5 (5 - 1)( 5 + 1) t li - 2 sin t]
1jT
-5(5-1)(5+1)(5-2)-
1)(5 + 1)TIi -
sli-3
sinsds
Chapter 5
412
+ (1- ~) [t 8 cost -15(15 - 1)(15 - 2)
lT
Jt8-I
sint - 15(15 _1)t 8- 2 cost]
s8-3 cos sds
and consequently,
Q(t)
T-too
urq(u)duds
t ii cos t
+ ",(t)
for all
t:::: to,
where for
",(t)
11 1
lim -T
-JtO-1sint-J(J-1)tO- 2 cost-J(J-1)(J-2)
Obviously, limHOO ",(t) =
00
t:::: to,
sO-3cos sds.
o.
Now let us consider an arbitrary continuous function k(t) on [to, 00)

with limHoo k(t) = O. Furthermore, let c(t) be defined by c(t) =
",(t) + k(t). Since limHoo c(t) = 0, we can choose a tl :::: max{to, 1} so
that c(t):::: -1/(2Y2) for all t:::: t l . Moreover, we consider an integer
N such that 27r N - (7r /4) :::: tl' Then for any integer n:::: N, we have
to cost 2: cost:::: 1/Y2 for every t E [2n7r - (7r/4), 2n7r + (7r/4)J. Thus, it
follows that
00
to
-[Q(s)
s
00
to
-1
L
8
+ k(s)J 2 ds
- [SO cos s + c( s)] ds::::

S
00
n=N
2n 1l'+(1l'/4)
2n1l'-(1l'/4)
-ds
S
2n7l'+(1l'/4)
00
n=N
1L In
= -
00
n=N
2n1l'-(1l'/4)
2]
1+-8n - 1
-1 ( v21
M -
2v2
)2ds
00.
Hence, in view of Theorem 5.2.8, equation (5.2.54) is oscillatory.

We note that for every t:::: to, one can obtain
Q*(t) =
t Q(s)ds
lto
- 315(15 - 1)
ft
ho
= t 8 sin t
+ 2Jtii - 1 cos t - tg sin to -
sO-2 cos sds - 15(15 - 1)(15 - 2)
it~ 1s
00
2Jtg- 1 cos to
u li - 3 cos ududs
and hence the function Q*(t) is unbounded. Therefore, Theorem 5.2.8 can
be applied to equations of type (5.1.6) without any additional hypothesis
that Q*(t) is a bounded function.
413

The following oscillation criterion is an extension of Theorem 5.2.8.
Theorem 5.2.9. Suppose condition (F) holds, and let p(t) E C 2 ([to, (0),
JR+) be such that for some constant c> 0,
(5.2.55)
iT is
and
lim -1
T->oo
to
to
rJ3(u)
(-) q( v,)dv,ds
a u
Define
1
lim -T
T->oo
exists in
IT IStau
rJ3(u) q(u)dv,ds
- (-)
JR.
(5.2.56)
t 2: to
for
and, suppose that

lim inf Q1 (t) =
- qo >
t ..... oo
qo
00,
is a positive constant.
Hfor every k(t) E C([to,oo),JR) with limHook(t)
001
ta
-;-[Q1(S)
+ k(sWds
(5.2.57)
= 0,
(5.2.58)
00,

for t 2: to Define w(t) = p{3(t)F(x(t)), t 2: to. Then as in

Theorem 5.2.6, (5.2.42) holds for every t 2: To 2: to. Thus, for any T, t
Proof.
x(t) =I-
To,
with T 2: t 2:
we have
(1 _i)
w'(t)
T
..!:..jT
IS p{3(u)
q(v,)dv,ds
T
a(v,)
1
+ T1 jT IS -gl(v,)w(v,)dv,ds + T1 jT jS w(u)
[w'(u) - hI (u)w(uW
_ w(T)
T
+ w(t) +
T
x f'(x(u))F(x(u))duds
and hence for every t 2: To,
w'(t)
1
lim -T
T ..... oo
00
Il
T
rJ3((u)) q(u)duds + lim wT(T)

tau
T ..... oo
+I
t
OO
-
g1 (u)w(v,)dv,
w(v,) [w'(v,) - h1(U)W(UW f'(x(u))F(x(u))du,
which shows that
. w(T)
11m
-- <
T->oo
00,
(5.2.59)
414
Chapter 5
roo -gl(u)w(u)du
<
(5.2.60)
00
JTo
and
00
To
-(-) [w'(u) - hI (u)w(uW f'(x(u))F(x(u))du <

W U
Now, by setting for t 2: To,

G1(t) =
100
t
-gl(s)w(s)ds+
100
t
(5.2.61)
00.
w(s) [w'(s)-hl(s)w(s)ff'(x(s))F(x(s))ds,
we find
w'(t) = Ql(t)
lim WT(T)+G(t)
forevery
'1'--+00
t2:To.
(5.2.62)
Next, we define
k(t)
weT)
T
{
lim weT)
'1'--+00
T
hm - -
T--+oo
wet) .
If t > To
t
+ Gl(To) _ w(To)
To
+ Gl(t)
- -
and observe from (5.2.59), (5.2.60) and (5.2.61) that k(t) E C([to, 00), JR.)
with limt--+oo k(t) = O. Moreover, in view of (5.2.59), there is a constant
K > 0 such that
we;:)
for all
T 2: To.
(5.2.63)
Also, since liminft--+oo[Ql(t) + k(t)] = liminft--+oo Ql(t)

there exists a constant f-t > 0, f-t ~ qo such that
Ql(t)
+ k(t) 2: -
f-t
for all
-qo
2: To.
>
-00
(5.2.64)
Now using (5.2.62), (5.2.63) and (5.2.64), we obtain for t 2: To,
rt _(1) [w'(s) _ hl(S)W(S)]2 f'(x(s))F(x(s))ds
JTo w s
2: K(/- (3)
! ~[w'(s)
rt
- hl(S)W(sWds
(3
1 [
K(l- (3) JTo :; Ql(S)
K(/- (3)
: ~
(T)
+ )~oo w T + G1(s) -
{[Ql(S) + k(s)] +
D-
h1(s)w(s)
2
hl(S)] w(S)} ds
]2
ds

{3
[ {t 1
K(l- (3) iTo -.;[Q1(S)
[1
>
K(:-
(t
+ k(s)]2ds + 2 iTo [Q1(S) + k(s)]
h 1(S)]
-s- w(s)ds + iTo
s2 -
(3)
[1
h 1(s)]2 2
-s- w (s)ds
S2 -
[h~ ~[Q1(S) + k(S)]2ds -
;::: K(:-{3) [,(
415
2/L
h~
[s12 - h 1;S)] W(S)dS]
~[QI(S)+k(s)fdS-2/LC,( -91(S)W(S)dS]
and consequently, because of (5.2.60) and (5.2.61), we have
(= ~[Q1(S) + k(s)]2ds::::: (TO ~[Q1(S) + k(s)]2ds + 2/LC t -91(S)W(s)ds

ito S
ito S
iTo
K(l - (3) {= 1
+
(3
iTo w(s) [w (s) - h1(S)W(s)]2 J'(x(s))F(x(s))ds < 00,
I
For the special case when a(t) = 1, p(t) = 0 and p(t) = t, we find
h ()
1
= f!..
t
[1 +!'(x)F(x)
f'(X)F(X)]
91(t)
and
and hence, condition (5.2.55) is satisfied.

immediate result.
(32
t2
[1~ -
1]
1 - f'(x)F(x)
Thus, we have the following
Corollary 5.2.12. Suppose condition (F) holds, and
1 (T
lim -T
r uf3q(u)duds
T-+= ito ito

Define
Q2(t)
1
lim -T
T-+=
iTt is
exists in
uf3 q(u)duds
for
ffi..
t:;::: to
and suppose that liminft-+= Q2(t) > -00. If for every k(t) E C([to, 00),
ffi.) with limt-+= k( t) = 0,
00,

x"(t)
+ (t~ sin t)f(x(t))
0,
t:;::: to> 0
(5.2.65)
Chapter 5
416
where f E C(JR., JR.) satisfies condition (F) and .A is a real number.

From Corollary 5.2.2 with p(t) = 0, we find that (5.2.65) is oscillatory
if .A > 1 - (3. Also, from Corollary 5.2.6, we observe that (5.2.65) is
oscillatory when - (3 < .A :::::: 1 - (3. Hence, equation (5.2.65) is oscillatory
if .A> -(3. Now, Corollary 5.2.12 guarantees that (5.2.65) is also oscillatory
for .A = -(3. Indeed, in this case for t::::: to, we have
Q2(t) =
lim
T-+oo
ITI
uf3 q(u)duds =
lITIs
lim -T
T-+oo
sinududs = cost,
and consequently, liminft-+oo Q2(t) = -1 > -00. Moreover, if k(t) E

C([to, (0), JR.) is an arbitrary function with limt-+oo k(t) = 0, then we
can choose a number Tl ::::: to such that k(t)::::: -1/(2)2) for all t::::: T 1 .
Furthermore, let N be an integer with 2N-rr - (-rr/4) ::::: T 1 . Then, it
follows that
00
12n11"+(11"/4)
n=N
2n11"-(11"/4)
1L
"8
00
n=N
1
1]2
-1 [ds
)2 2)2
S
2]
In 1 + 8n _ 1
[
00.
Thus, all conditions of Corollary 5.2.12 are satisfied, and therefore, equation
The following result not only extends Theorem 5.1.6 to more general
equations of the type (5.1.1), but also removes the condition Q(t)
!too q( s )ds :::::
for t::::: to.
Theorem 5.2.10. Let condition (F) hold, and Q(t) =

for t::::: to and satisfy
it
lim
sf3- 1 Q(s)ds
t-+oo to
00,
ft q(s)ds exist
OO
(5.2.66)

Proof. First, we shall show that if f(x) satisfies condition (F), then
lim f(x)
x-+oo
= 00.
(5.2.67)
In fact, from condition (F) and the definition of F(x), we have F'(x) =
l/f(x) and
f'(x) > 1 F'(x)

f(x)
~ F(x} ,
(5.2.68)
417
where lie = (31(1- (3). Integrating (5.2.68) from Xo to x, we find
and hence
( f(x)
f(xo)
)C
F(x)
F(xo)
> --
(5.2.69)
Applying condition (F) to inequality (5.2.69), we obtain
cJ'(x)(f(x))C 2: F(lx ) (f(x))C 2: (f(xo))C

F(xo) .
(5.2.70)
Integrating (5.2.70) from Xo to x, we get
which shows that f(x) -+ 00 as x -+ 00. A similar argument applies for

x < Xo and results in f(x) -+ -00 as x -+ -00. This proves (5.2.67).
So, if x(t) is a nonoscillatory solution of (5.1.1), then by Lemma 4.1.3, it
satisfies the integral equation
x'(t)
f(x(t))
Q(t) +
oo
x'(s)
f(x(s))
)2 f ,(x(s))ds,
(5.2.71)
In particular, we have
00
x'(s)
f(x(s))
to
)2 J'(x(s))ds
<
(5.2.72)
00
from which we shall show that x(t) has the following asymptotic behavior
lim F(x(t))
t-+oo
= o.
(5.2.73)
To prove (5.2.73), we apply the Schwartz inequality to estimate F(x) as

follows
F(x(t)) - F(X(t1))
<
li
tl
x'(s)
f(x(s)) ds
[i t(f~X(:))
'())2
tl
]1/2
J'(X(S))dSJ
[it
tl
f'(x(s)) ds
]1/2
(5.2.74)
418
Chapter 5
for t 2': tl 2': to. In view of (5.2.72), we choose tl sufficiently large so that
for any E > 0,
00 (
x' ( s ) ) 2 I
E
f (x(sds <
f(x(s
t,
4'
Using the above inequality and condition (F) in (5.2.74), we obtain
F(x(t
F(X(tl
;E 1, F(x(sds )1/2
fo (
(5.2.75)
Now suppose F(x(t E .cl(tl' 00). Then, F(x(t is bounded by (5.2.75),

and hence (5.2.73) is satisfied. Otherwise, we can choose a t2 2': tl so that
;E 1, F(x(sds )1/2
~ fo (
F(X(tl
which together with (5.2.75) yields
F(x(t
fo
(1:
for
F(X(SdS) 1/2
(5.2.76)
Integrating (5.2.76) from t2 to t, we get

(
1,
F(x(sds
) 1/2
1,t?
F(x(sds
) 1/2
~ fo (t _ t2) < fo t
22'
(5.2.77)
Once again, we can choose t3 2': t2 so that
(t 2 F (X(sds <
itl
~Et2
t2': t3.
forall
Thus, from (5.2.76) and (5.2.77), (5.2.73) follows immediate. Now, we define wet) = t iJ - l F(x(t)), t 2': to. Then for t 2': to, we have x'(t)/ f(x(t =
tl-iJw'(t) + (1 - (3)c;jw(t) and
I
f (x(t
x'(t)
f(x(t
)2
f'(x(tF(x(t [t1-(3 [w'(tW + 2(1-{3)C iJ w'(t) + (1 - (3)2C l-iJw(t)]

wet)
1
[w ' (t)J2
2': _t l - iJ
+ 2)..CiJW' (t) + (3(1- (3)C l -iJw(t).
c
wet)
(5.2.78)
Integration of the second term on the right-hand side of (5.2.78) from t
to T 2': t 2': to yields
w~) +{31TW}+s~dsl.
2{31 T s-iJw'(s)ds = 2{3 [W(']) _

t
T"
t"
S"
(5.2.79)
Oscillation theory for sublinea.r differential equations
419
In view of (5.2.73) and the fact that w(t) is nonnegative the limits of
the two integrals in (5.2.79) exist and can be infinite. Noting (5.2.72) and
(3 < 1, we can integrate both sides of (5.2.78) and deduce that the limit
on the left-hand side of (5.2.79) must be finite. Hence, the integral on the
right-hand side of (5.2.79) is likewise finite, and therefore, we can define an
energy function
E(t)
w(t) - (1
+ (3)tf31O ~~? ds,

S
t 2 to.
(5.2.80)
Recall that x(t) is a solution of (5.1.1) and satisfies the integral equation
(5.2.71). Computing E'(t), we find
x'(t)
E'(t) = t f3 - 1 _
_
f(x(t))
+ ({3 - 1)t 1- f3 F(x(t)) -
+(1 + (3)tf3 ~~?
(3({3 + 1)tf3 - 1
00
w(s)
ds
SM1
t 2 to.
for
(5.2.81)
Combining the second and the last terms in (5.2.81) and using (5.2.71), we
obtain
E'(t)
tf3-1 [Q(t)
-(3(1
00
x'(s)
f(x(s))
)2 f'(X(S))dS] + 2{3w(t)
+ (3)tf3 - 1 1O w(S)1 ds,

sf3-
t 2 to
(5.2.82)
where (3 = 1/(1 + c) < 1 and c = (1 - (3)/(3 > O. Furthermore, we
apply (5.2.73) and combine (5.2.78) and (5.2.79) with (5.2.82), to obtain
for t 2 to,
E'(t) 2 t f3 - 1Q(t)
1 f3 - 1
+ _t
c
00
Sl-A [w'(s)J2 ds 2 t f3 - 1Q(t).

w(s)
Integrating this inequality and using condition (5.2.66), we get limHoo E(t)
= 00. From the definition of E(t) in (5.2.80), we also deduce that
limHoow(t) = 00. Define tn = sup{t : w(t) ::; n}, which satisfies
limn-+oo tn = 00 and also t 2 tn implies w(t) 2 n = w(t n ). Replacing
t in (5.2.80) by tn and using its definition, we obtain
Chapter 5
420
which contradicts the fact that limn-too E(t n )
proof.
= 00.
This completes the
Exrunple 5.2.11. Consider the sublinear differential equation
xl/(t) + f Y: 1 ['Y + 'Yvtsint + tvt cos t] Ix(t)I'Ysgn x(t) = 0,
(5.2.83)
where 0 < 'Y < 1, t 2': to> O. Let f(x) = Ixl'Ysgn x so that c = (l-'Y)h
and fJ = 'Y. Also,
Q(t)
1=
q(s)ds =
1
-[l+vtsint],
t'Y
t 2': to
It is easy to check that all hypotheses of Theorem 5.2.10 are satisfied and
hence equation (5.2.83) is oscillatory.
We note that Theorem 5.1.6 fails to apply to equation (5.2.83) since the
function Q(t) assumes negative values for all large values of t and satisfies
condition (5.1.25). Therefore, we conclude that Theorem 5.2.10 improves
Theorem 5.1.6.
x"(t) +
t~
G ~vtsint
+
+ tvt cos t] {lx(t)I- 1/ 2 + Ix(t)1 1/ 2 } x(t) = O.
(5.2.84)
We let
(5.2.85)
f;
so that F(x) =
du/ f(u) = 2 tan-1..jX. Clearly, condition (F) is satisfied
with fJ = 1/2, i.e., c = (1 - fJ) / fJ = 1. All conditions of Theorem 5.2.10
are satisfied and hence equation (5.2.84) is oscillatory.
It is interesting to note that f(x) in (5.2.85) satisfies both conditions
of sublinearity, i.e., (5.1.5) as well as of superlinearity, i.e., (4.1.5).
Next, we shall prove the following result.

Theorem 5.2.11. Let condition (5.1.5) hold and
min
{inf
f'(x) r fd(U) , inf f'(x) jX fd(U)}
x>o
J+ o
x<o
-0
> O.
(5.2.86)
Furt1J.er, let there exist p(t) E C 1([to,00),1R+) such that
(a(t)p(t))':::; 0,
~(t)=
a(t)p'(t)
= p(t)p(t)
() ()ds-too
toasps
as
for
t-too.
t 2': to, and

(5.2.87)
421

If
lim sup -1()
~ t
t-400
and
liminf - 1
t-400
tm
it
to
it
to
( )1 ( )
asp s
[is
to
p(u)q(u)du
(t - s)mp(s)q(s)ds > -
ds
00
for some integer
00
(5.2.88)
m21,
(5.2.89)

x(t) > 0 for t 2 to. We define w(t) = Iox(t) du/ f(u) for t 2 to. Then from
(5.1.3) it follows that w(t) satisfies the second order nonlinear differential
equation
Proof.
(a( t)p( t)w' (t))'
-p(t)q(t)
+ (a(t)p'(t) -
x'(t)
p(t)p(t)) f(x(t))
-a( t)p( t)( w' (t))2 f' (x( t))

-p(t)q(t) - a(t)p(t)(w'(t)J2 f'(x(t)),
It:
Now, with respect to the function I(to, t) =

we shall consider two mutually exclusive cases.
t 2 to
(5.2.90)
a(s)p(s)(w'(s))2 f'(x(s))ds
Case 1. I(to, 00) < 00. In view of (5.2.86), there exists a positive constant
a such that
f'(x(t))w(t)
ret)
f'(x(t)) J+o
We shall show that
du
f(u) > a
w(t)
hm -(-)
t-400 ~
For this, let

so that
for all
t 2 to
(5.2.91 )
(5.2.92)
O.
> 0 be an arbitrary number. Then we can choose a tl 2 to

(5.2.93)
By Schwartz's inequality for every t 2 tl, we have
w(t) - W(tl) ::;
11: w'(s)dsl
::; (I(tl' t))1/2 (.:r(tl, t))1/2 ,
where .:r(tl, t) = Itt11/(a(s)p(s)f'(x(s)))ds,

(5.2.93), we find
w(t) ::; w(iI) +
~ (.:r(tl, t))1/2
and consequently, by using
for all
t 2 tl'
(5.2.94)
Chapter 5
422
If J(tl, (0) < 00, (5.2.94) ensures that wet) is bounded, and so (5.2.92)
holds. Thus, we restrict our attention to the case when J(tl, (0) = 00. In
this case, we can choose a
> h so that W(tl)::; (-JfJi/2) (J(tl,t1/2
for all t ~ t2. Thus from (5.2.94) it follows that wet) ::;
(J(tl,
for all t ~ t 2 , and so, by using (5.2.91), we have
t2
t1/2
..;w.
(5.2.95)
or
{
L
t
}-1/2
w(s)
a(s)p(s) ds.
Integrating this inequality from t2 to t

2
vE
w(t)::;
~
for all
t2, we obtain
() d }1/2
ws
{i a(s)p(s) ds }1/2 - {it? a(s)p(s)
t
()
WS
t,
tJ
~ t 2
<
-
vE
to
a(s)p(s)
ds.
Hence, by setting
max { t2,
t3
for t
{l
-IE
t2
t,
w(s)
a(s)p(s) ds
}1/2}
t3, we obtain
Thus, from (5.2.95) it follows that wet) < E~(t) for all t ~ t3. As E > 0
is arbitrary, the proof of (5.2.92) is complete. Now, from equation (5.2.90),
p(u)q(u)du = C l - a(t)p(t)w'(t) -I(to, t), where
for t ~ to, we have
C1 = a(to)p(to)w'(to). Hence, for every t ~ to, we find
It:
[1:
P(U)q(U)dUf
[Cl
a(t)p(t)w'(t) - I(to, t)]2
< 3cl + 3[a(t)p(t)w'(tW + 3 (I(to, t2

< C2 + 3[a(t)p(t)w'(tW,
where C 2 = 3Ci
+ 3 [I(to, (0)]2.
Next, we make use of (5.2.91) to obtain
423

for every t
to,
~tt) 1: a(s)lp(s) [1: P(u)q(U)dUf ds
it
s::
3 to a(s)p(s)(w'(s))2ds
C2 + ~(t)
s::
C2
+ (3/(;))
~
a(s)p(s)(w'(s))2 f'(x(s))w(s)ds
.Jto
< C2 + (3/(a)) [max w(s)]

~ t
and consequently for t

1
~(t)
to:::;s:::;t
t a(s)p(s)(w'(s))2 f'(x(s))ds
.Jto
to,
itto a(s)p(s)
1
[isto p(u)q(u)du] 2 ds s::
C2
+ C3
(~ttJ [t~~~t w(s)] ,
(5.2.96)
where C 3 = (3/a)I(to, (0) > O. But, by (5.2.92) we can choose T ~ to
so that w(t) s:: ~(t) for every t ~ T. Therefore, maxto<s<t w(s) s::
maXto:::;s:::;T w(s) + ~(t) for all t ~ to. Thus, (5.2.96) gives - -
for all t
to. Hence, we have
limsup C(l)
t-.oo
."
it
to
()1 ( )
asp s
[is
to
p(u)q(U)dU] ds
s::
C2 +C3 <
00,

Case II. I(to, (0) =
From equation (5.2.90) for every t
to, we have
(t - s)mp(s)q(s)ds + it (t - s)ma(s)p(s)(w'(s))2 f'(x(s))ds
-1t
+ 1t
it
to
00.
to
to
(t - s)m(a(s)p(s)w'(s))'ds
(t - to)a(to)p(to)w'(to) - a(t)p(t)w(t)
to
(a(s)p(s))'w(s)ds
(t - to)ma(to)p(to)w'(to)
+m
to
for m
+ a(to)p(to)w(to)
=1
+ m(t -
to)m-la(to)p(to)w(to)
[-(m-1)(t-sr- 2a(s)p(s)+(t_s)m-l(a(s)p(s))'] w(s)ds

if m
424
::;
Chapter 5
(t - to)a(to)p(to)w'(to) + a(to)p(to)w(to) if m = 1
(t-to)ma(to)p(to)w'(to) + m(t-to)m- 1 a(to)p(to)w(t o) if m 2: 2.
Therefore, for all t 2: to,

1 i t(t - s)mp(s)q(s)ds ::; --;;
1 i t(t - s)ma(s)p(s)(w'(s)f f'(x(s))ds
-;;
t
~
t
~
to
,
m
to
+ 1- t
a(to)p(to)w (to) + t 1 - t
a(to)p(to)w(to).
() m
()m-1
(5.2.97)
But, since
lim
t-400
~
i t (t m
t
s)ma(s)p(S)(W'(S))2 f'(x(s))ds
to
I(to, (0)
00.
inequality (5.2.97) gives

lim - 1 i t (t - s)mp(s)q(s)ds
t-400
tm
to
= -
00,

For the special case when pet)
result.
= 0 and pet) = 1, we have the following
Corollary 5.2.13. Let conditions (5.1.5) and (5.2.86) hold, and a'(t) ::; 0
for t 2: to and ~l(t) = Itto ds/a(s) ----+ 00 as t ----+ 00. If
1
lim sup -(-)
t-400
and
liminf - 1
t-400
tm
it
to
6 t
it [is
to
1
-(-)
a s
to
(t - s)mq(s)ds > -
q(u)du
00
ds
00
for some integer
(5.2.98)
m 2: 1, (5.2.99)

Also, when a(t)
corollary.
= 1, pet) = 0 and pet) = 1, we have the following
Corollary 5.2.14. Let conditions (5.1.5), (5.2.86) and (5.2.99) hold, and
limsuP~lt
t
t-400
to
[is
to
q(U)dU] ds
00,
(5.2.100)
425
Remark 5.2.6.
1. The condition
lim sup -1
t
t--+oo
jt.jS q(u)duds
to
(5.2.101 )
00
to
implies (5.2.100) as an application of the Schwartz inequality.

2. The condition (5.2.99) holds if
liminf -1
t
t--+oo
jt IS q(u)duds >
(5.2.102)
00.
to. t.o
Now, we can state that equation (5.1.6) with 0 < --y < 1 is oscillatory
if either one of the following holds:
(II) conditions (5.2.101) and (5.2.102),

(12) conditions (5.2.100) and (5.2.102),
I1 l
-00 < liminf t--+CXJ
to
to
I1t I
q(u)duds < lim sup t---+CXJ

t
to
to
q(u)duds ::; 00.

(5.2.103)
Next, we shall prove the following more general result for (5.1.3). It
extends the criterion (Id above.
Theorem 5.2.12. Let conditions (5.l.5) and (5.2.86) hold, and assume
that there exists a function p(t) E C 1 ([to,00),lR+) such that conditions
(5.2.87) and (5.2.89) are satisfied. If, for some integer n > I,
lim sup -1
t--+CXJ
tn
and
lim sup -1
t--+=
tn
It
(t - s)rlp(s)q(s)ds
(5.2.104)
00
to
It
(t - s)n-2~(s)ds
< 00,
(5.2.105)
to

Proof. Let x(t) be a nonoscillatory solution of (5.1.3), say, x(t)
t :::> to. As in Theorem 5.2.11, we define w(t)
J;(t) du/ f(u),

=
t :::> to which
a(t)p(t)w'(t), t :::>
to.
(5.2.106)
reduces (5.1.3) to equation (5.2.90). Next, we let z(t)

to. Then equation (5.2.90) becomes
,
z (t)
J'(x(t)) 2)
a(t)p(t) z (t
+ p(t
q(t) = 0,
t :::>
> 0 for
As in Theorem 5.2.11 we consider the two cases I and II. The proof of case
II is exactly the same as earlier and hence omitted. Now, for the case I
426
Chapter 5
we proceed as in Theorem 5.2.11, to obtain (5.2.91) and (5.2.92). Using

(5.2.91) in equation (5.2.106), we find
z'(t)
+ w7t) Z2(t) + p(t)q(t) ::;
for
t 2:: to
From (5.2.92), there exists a
tl
2:: to such that for all
w(t)
s:
~(t)
for
(5.2.107)
2:: tl.
2::
tl
we have
(5.2.108)
Using (5.2.108) in inequality (5.2.107), we get
s:
z'(t) + ~~) z2(t) + p(t)q(t)
for
t 2::
(5.2.109)
tl.
Multiplying (5.2.109) by (t-s)" and integrating from tl to t, we obtain
or

1
-:;;
t
it
t,
(t-s)np(s)q(s)ds
s:
t)t
1 - .2
n2) -:;;t1 it ~(s)(t-s)n-2ds.
Z(tl)+ ( -
40
t,
Taking lim sup on both sides of the above inequality as t -+ 00 and

applying condition (5.2.105), we obtain a desired contradiction to condition
(5.2.104).
In Theorem 5.2.12, if (5.1.5) and (5.2.86) are replaced by condition (F),

then we get the following result.
Theorem 5.2.13. Suppose condition (F) holds with (3 > 0 and c = (1(3)/ (3. Moreover, assume that there exists a function p(t) E Cl ([to, 00), :JR+)
such that conditions (5.2.87) and (5.2.89) are satisfied. If for some integer
n> I,
1
lim sup -:;;
t400
it
to
n2 c
] ds = 00, (5.2.111)
(t - st- 2 (t - s)p(s)q(s) - -~(s)
4
427


x(t) > 0 for t ~ to. Proceeding as in Theorem 5.2.11, we obtain the
Proof.
inequality (5.2.110) which takes the form
1~ (t -
s)n-2
[(t - s)2p(s)q(s) -
~~ ~(s)] ds
(1 _t; ) n z(t
::;
Dividing both sides of the above inequality by tn,

both sides as t -t 00 and noting that 0: = 11 c,
contradiction to condition (5.2.111).
1 ).
taking lim sup on

we obtain a desired
The following results are immediate consequences of Theorems 5.2.11

and 5.2.12.
Corollary 5.2.15. Let conditions (5.1.5) and (5.2.86) hold, or condition
(F) holds with (3 > 0, a'(t) ::; 0 for t ~ to and 6(t) =
dsla(s) -t 00
as t -t 00. If condition (5.2.99) holds and for some integer n > 1 either
It:
it
it
1
limsup--;t
t-->oo
and
1
lim sup --;-
t-->oo
or
limsup :
t-->oo
to
to
to
(t-s)nq(s)ds
6(s)(t - s)n- 2ds <
(5.2.113)
00,
(t-s)n-2 [(t-S)2 q(S) - n 2c 6 (S)] ds

4
where c = (1 - (3)1(3,
(5.2.112)
00
00,
(5.2.114)
Corollary 5.2.16. Let either condition (F) hold with (3 > 0, or conditions (5.1.5) and (5.2.86) hold. If conditions (5.2.99) and (5.2.112) hold,
Corollary 5.2.15 follows from Theorem 5.2.11 (Theorem 5.2.12) by letting pet) = 1 and pet) = 0, while Corollary 5.2.16 follows from Corollary
5.2.15 by letting aCt) = 1 and noting that for n> 1,
lim -1
t-->oo
tn
it
t1
set - s)n- 2ds =
n n- 1
) <
00.
Remark 5.2.7.
It is known that (5.2.112) alone is sufficient for the
oscillation of sublinear equation (5.1.6). Also, from Theorem 5.2.12, we
note that the condition
lim -1
t-->oo t n
lt
to
(t - s)nq(s)ds =
00
for some integer
n> 1
(5.2.115)
Chapter 5
428
alone is sufficient for the oscillation of (5.1.6) with 0 < 'Y < 1.
We also note that conditions (5.2.112) and (5.2.115) can be improved
by allowing n to be any real number > 1 and the proof remains the
same.
Remark 5.2.8. In Theorem 5.2.12 the restriction of selecting the weight
(it:
function pet) = exp

(p(s)/a(sds) can be removed. This requires
only slight modification in the proof. The details are left to the reader.
(~Xl(t)' + 6~2XI(t)+ [-CI/3sint+~C4/3(2+Cost)]

x
= 0, 0 < 0: < 1,
{lx(t)I" sgn x(t) + x(t)}
t 2: to
= 'IT /2.
(5.2.116)
Here, f(x) = Ixl"sgn x + x, f'(x) = o:lxl"-l + 1 > 0 for all x f. 0, and
F(x) = fJxl du/(u" + u). To estimate F(x) from below, we observe that
F x
()
fixi ~ >
u'" + u -
fix
10
10
and
F(x) =
Therefore, for x
1xI
f.
-du- >
u'"
+u
11
0
Ixll-"
2(1 - 0:)
-du =
1
2(1 - 0:)
2u
2u a
2u"
if
Ixl S; 1,
if Ixl 2: 1.
0 we have for Ixl S; 1,
J'(x)F(x) >
lxiI-a [o:l xl",-l + 1] >
2(1 - 0:)
2(1 - 0:)
and for Ixl 2: 1,

2(1 - 0:)"
Using condition (F), we find (3 = 0:/(2 - 0:).
Next, let pet)
holds with
{(t)
1t
(1f/2)
= t5/ 6 .
1
a(s)p(s)
Then for t 2: to =
ds
1t
(1f/2)
SI/6ds
'IT /2,
we find that (5.2.87)
= -7
6
e/6 -
'IT
7/6]
C)

Furthermore, for every t
1
2"
429
to,
itto (t - s)2 s5/6q(s)ds

~ it (t - s) [is u 5/ 6q (U)dU] ds
to
to
~ it (t - [is u 5/ 6 [_u- 3
to
to
2jt (t - it
(2 +
~ jt (t _ [Sl/2(2 +
t
> ~ it (t - s) [Sl/2 _ (~) 1/2]
2
t
s)
2"
t
s)
(11"/2)
(11"/2)
sin 1t +
d [u 1 / 2
s)
(11"/2)
1/
cosu)] ds
coss) _ 2
(11"/2)
~u-4/3(2 + cos u)] dU] ds
(~) 1/2] ds
2
ds
~tl/2 _ 2 (~)1/2 + ~ (~)3/2 ~ _ ~ (~)5/2 ~

15
t2
and consequently condition (5.2.111) is also fulfilled with n = 2.

Theorem 5.2.13 guarantees that equation (5.2.116) is oscillatory.
So,
5.3. Linearization of Sublinear

Oscillation Theorems
In this section we shall relate the oscillation of sublinear differential
equations to some linear second order equations. Our first result in this
direction is the following.
Theorem 5.3.1.
Let condition (F) hold, and suppose there exists a
function p(t) E C2([to, 00), JR+) satisfying (5.2.24). If
lItIS
liminf t-400
and for every constant k
to
to
rI(u)q(u)duds > - 00
(5.3.1)
> 0, the linear second order equation
(tv'(t' + C(t)v(t)
(5.3.2)
0,
C(t) = krl(t)q(t) - {J2 t (p'(t)2 + ~ (tP'(t)'

4
p(t)
2
p(t)
for all large
t,
Chapter 5
430

x(t) of. 0 for t;::: to, and let w(t) = pO(t)F(x(t)). Then as in Theorem
Proof.
5.2.5, we have
w(t)
hmsup- <
t--+oo
(5.3.3)
00.
We also have
w
jJ-I'(
(t)
(3p
x'(t)
+ PO( t) f(x(t))
'
((
(t)p t)F x t))
or
z(t),
t;::: to
and
z'(t)
(pO(t)
f~S;)))'
'(t)
Pt
-~(t)q(t) + (3~() z(t) - -(-)z2(t)f'(x(t))F(x(t)),
t;::: to.
w t
(5.3.4)
From condition (F) and (5.3.3), we find that there exist a tl ;::: to and a
constant C1 > 0 such that
f'(x(t)F(x(t)) ;::: 1 ~ (3
w(t)
for
t;::: tl
+ (3~(t) z(t) - Ciz2(t),
t;::: t]
and
S; CIt
(5.3.b)
z'(t)
where
S;
pO(t)q(t)
/(t)
= (3!(cI(1 - (3)). Thus, for t;:::
( z(t) _
~tP/(t))' <
2c p(t)
-pO(t)q(t) _
-~t
[z(t) -
tl
we have
(tP'(t))/
p(t)
2c
~tP/(t)]2
2c p(t)
,
+ (32 t (p/(t))2
t>
-
p(t)
4c
tl
Now, let
y(t) = z(t) -
~tP'(t),
2c p(t)
t;::: tl
so that
y'(t)
S;
[pO(t)q(t) - (32 t (p'(t))2

4c
p(t)
+~
2c
(tP'(t))'j_ ~y2(t),
p(t)
t
t;::: tl.
(5.3.6)
Finally, applying Lemma 2.2.1, we find that equation (5.3.2) is nonoscillatory, which is a contradiction. This completes the proof.
431

Corollary 5.3.1.
Let condition (F) hold, and suppose there exists a
function p(t) E C 2([to, (0), JR+) such that p'(t) ~ 0, p"(t) :::::: 0 and
(tp'(t)/p(t))' ~ 0 for t ~ to. If condition (5.3.1) holds and for every
constant k > 0 the second order linear equation
(tv'(t))'
+ [k tf3 (t)q(t) - ~2] v(t)

0 for t ~ to. Proceeding as in Theorem 5.3.1, we obtain inequality
(5.3.6). From the hypotheses this inequality becomes
Proof.
x(t)
-=F
Y (t) ::::::
132-1
- tf3(t)q(t) - 4c t
(t-p(t)(t) ) 2 p'
-y (t)
t
and now by Remark 5.2.2, we find
Theorem 5.3.1 can be extended to more general equations of the type

(5.1.3). In fact, we have the following result.
Theorem 5.3.2. Let condition (F) hold, and let p(t) E C2([to,00),JR+),
a(t) E C1([to,00),JR+) and p(t) E C1([to,00),JRo) so that condition
(5.2.14) be satisfied with the functions g(t) and h(t) as in (5.2.9) and
(5.2.10) respectively, (p!3(t)p(t)/a(t))':::::: 0 for t ~ to and
lit is
liminf t-too
If for every constant k
(tv'(t))'
to
to
p!3(u)
- (-) q(u)duds > - 00.
(5.3.7)
a u
> 0, the second order linear equation
[k~(~~) q(t) + ~(tP(t))' - ~tP2(t)] v(t)
(5.3.8)
P(t) =
a(t) (tf3(t))' p(t)

p!3(t) a(t)
- a(t) ,
(5.3.9)
Chapter 5
432

x(t) -=J. 0 for t;:::: to and let wet) = pf3(t)F(x(t. Then as in Theorem
Proof.
5.2.5, we obtain (5.3.3). Now, it follows that
pf3(t) a(t)x'(t)
aCt) f(x(t
w'(t) - (3p'(t) wet)

pet)
z(t),
t;::::to
and
Z
pf3(t)
[aCt) (pf3(t) , pet)]
1 2
,
(t) = - aCt) q(t)+ pf3(t) aCt) - aCt) z(t)- wet) z (t)f (x(tF(x(t.
(5.3.10)
From condition (F) and (5.3.3) there exist a tl ;:::: to and a constant CI > 0
such that (5.3.5) is satisfied. Using (5.3.5) in equation (5.3.10), we find
where
C
(3
CI
(1 _ (3)
Thus, for all t;:::: t l
and
,
pet)
aCt) (pf3(t) ' pet)

pf3(t) aCt)
- a(t)'
t;:::: to
we have
pf3(t)
1
,1
2
--q(t) - -(tP(t + -tP (t)
aCt)
2c
4c
-~
[z(t) -
;c
tP(t)f,
t;::::
tl
or
Y, (t) :::;
- [pf3(t)
-(-) q()
t
a t
1 (tP ('
+ -2
t c
1
2')]
C 2( t,
)
-tP
(t - -y
4c
t
t;::::
tl
where yet) = z(t) - (1/2c)tP(t), t ;:::: tl' The rest of the proof is similar
to that of Theorem 5.3.1 and hence omitted.
Next we present the following linearization result.

Theorem 5.3.3. Let conditions (5.1.5) and (5.2.86) hold, and let pet) E
C I ([to,oo),1R+) so that (5.2.87) and (5.2.89) be satisfied. If for every
constant k > 0, the second order linear equation
(a(t)p(t)~(t)v'(t'
+ kp(t)q(t)v(t)
(5.3.11)
433

x(t) > 0 for t 2: to, and let w(t) = J;(t) du/f(u), t 2: to. Then as in
Theorem 5.2.11, we find that (5.1.3) is reduced to equation (5.2.90). Next,
we consider the cases I and II as in Theorem 5.2.11. The proof of case II is
exactly the same and hence omitted. For case I, we proceed as in Theorem
5.2.11 to obtain (5.2.91) and (5.2.92). Now, as in Theorem 5.2.12, we let
z(t) = a(t)p(t)w'(t) so that equation (5.2.90) takes the form
Proof.
z'(t) = - p(t)q(t) _
or
y'(t) = p(t)q(t)
()1 ( ) z2(t)f'(X(t)),
a t p t
+ a(t)p(t)y2(t)f'(x(t)), t
2: to
where y(t) = -z(t) for t 2: to. If we let c(t) = 1/ f'(x(t)), t 2: to then

the Riccati equation y'(t) = p(t)q(t) + (c(t)y2(t)/a(t)p(t)) has a solution
on [to, (0). It is well-known that this is equivalent to the nonoscillation of
the second order linear differential equation
(5.3.12)
Now, from (5.2.91) and (5.2.92) there exist a tl 2: to and a constant CI > 0
such that f'(x(t))w(t) 2: a and w(t) :::; CI~(t) for t 2: tl' Thus, it follows
that c(t) 2: a/(cI~(t)) = C2/~(t) for t 2: t l , where C2 = a/cI > O.
Finally, applying the Picone-Sturm theorem to the equation
C12 a(t)p(t)~(t)V'(t))' + p(t)q(t)v(t)
we conclude that it is nonoscillatory, contradicting the fact that (5.3.11) is
oscillatory. This completes the proof.

The following result is immediate.
Corollary 5.3.2.
Let conditions (F) and (5.2.99) hold. If for every
constant k > 0, the second order linear equation (ty'(t))' +kq(t)y(t) = 0
5.4. N onoscillation Theorems for

Sublinear Differential Equations
Here, we shall discuss the nonoscillatory behavior of the Emden-Fowler
equation (5.1.6) with 0 < 1 < 1. It is well-known that (5.1.6) has a
434
Chapter 5
nonoscillatory solution if q(t) satisfies
q(t) 2: 0 for
t 2: to
and
00
s'Yq(s)ds < 00.
to
(5.4.1)
It:
Condition (5.4.1) in particular implies that limt--+oo

q(s)ds exists and
is finite when q(t) is nonnegative. The following theorem extends this
result to the case when q(t) E C([to,oo),JR).
Q( t)
00
q( s )ds
t 2: to.
exists for all
(5.4.2)
If there exists a function F(t) E C1([to,00),IRo) such that
IQ(t)1 ::; F(t)
for all large
t,
(5.4.3)
where F(t) = O(t-'Y) as t --+ 00, and
00
s'YIF'(s)lds < 00,
(5.4.4)
to
then equation (5.1.6) has a nonoscillatory solution.

Proof. Let xm(t) be a solution of equation (5.1.6) satisfying xm(l) =
0, x~(1) = m where m is a positive number. We claim that when m
is large enough, x~ (t) > 0 for all t > 1 and so x(t) is nonoscillatory.
Suppose that x~(t) = 0 for some t > 1. Let 6 be the smallest of
such t. Let 6 be the smallest of all those t for which x~(t) = 2m.
(If no such t exists, let 6 = 00). Finally, let

[1,~], 0 < x~(t) < 2m. Thus, it follows that
o<
At t
=~,
Xm(t) < 2mt,
min {6, {2}. Then on
~ =
E [1, {].
(5.4.5)
we have either
x~(~) =
(if ~ = {I)
or
Integrating (5.1.6) from 1 to t E
x~(t)
x~({) =
[1,~],
-it
2m
(if ~ = 6).
(5.4.6)
we find
q(s)x'Y(s)ds.
(5.4.7)
We now proceed to estimate the integral in (5.4.7) as follows
lit q(S)X~(S)dSI
I[Q(l) -
Q(t)]x~(t) +
it
< [2IQ(1)1 + IQ(t)l]x~(t) +
[Q(s) -
Q(l)](x~(s))'dsl
it IQ(s)l(x~(s))'ds
(5.4.8)
435
since xm(t), x~(t) > 0 on [1, ~).

Next we integrate the last integral in (5.4.8), to obtain
1ft Q(s)(x~(s))'dsl
s;
<
ft P(S)(X'(s))'ds
P(t)x7r,(t) + ft 1F'(s)lx7r,(s)ds.
(5.4.9)
Since Q(t) -+ 0 as t --+ 00, Q(t) is bounded on [1, .;), i.e., IQ(t)1 s;
kI, t E [1,';) where k is a constant. By assumption there exists a constant
k2 such that It' P(t)1 S; k2, t E [1,.;). For t E [1, .;), we also have from
(5.4.5),
ft 1F'(s)lx7r,(s)ds
S;
(2m)'
ft s'IF'(s)lds
k3(2m)',
S;
(5.4.10)
for some constant k3 > O. Using (5.4.9) and (5.4.10) in (5.4.8), we find
1ft q(S)X;n(S)dSI
S;
[3k1
+ k2 + k3](2m)' = K(2m)'.
(5.4.11)
Using (5.4.11) in (5.4.7), we obtain

m ~ K(2m)' S; x~(t) S; m
+ K(2m)'
for all
For m > (2' K)l/(l-,), we have in particular 0

contradicts (5.4.6).
<
t E [1,';].
x~(~)
< 2m. This
xl/(t)
+ (t A sin t)lx(t)I'
sgn x(t)
= 0,
< 'Y < 1,
t 2: to > 0 (5.4.12)
where ,\ <~,. We find that IJtCXJ q(s)dsl S; ktA, t 2: 1 where k is a

positive constant. Let P(t) = ctA, where c is a positive constant. All
the hypotheses of Theorem 5.4.1 are satisfied, and hence equation (5.4.12)
is nonoscillatory.
xl/(t)
+ [t~(lnt)l-'sint]
Ix(t)I' sgn x(t)
0,
0 < 'Y < 1,
t 2: to> 0
(5.4.13)
where p, S; ~2. We find that P(t) can be taken to be a multiple of
(lnt)l-'/t. If P(t) E C1([to,00),lRo) and P'(t) 2: 0 for t 2: to such that
fCXJ S,-l P(s)ds
< 00,
(5.4.14)
Chapter 5
436
then F(t) satisfies the hypotheses of Theorem 5.4.1, i.e., F(t) = O(t-/,)
as t -+ 00, and (5.4.4) holds. To see this we apply integration by parts,
to obtain
1 F(t)
-t'
'"Y
/,t -s/'[-F'(s)]ds /,t

1
'"Y
S/,-l
F(s)ds
1
+ -F(I).
'"Y
(5.4.15)
Since the right-hand side of (5.4.15) is bounded, by (5.4.14) each of the

terms on the left-hand side is also bounded for all t. It can be shown by
continuity argument that the theorem still holds if F(t) satisfies (5.4.14)
but no continuity requirement is assumed on F.
Next, we shall consider equation (5.1.6), where q(t) E C([t o, 00), JR+)
and is locally of bounded variation on [to, 00). Thus, q( t) admits a
Jordan decomposition q(t) = q+(t) - q-(t), where q+(t) and q-(t) are
continuous nondecreasing functions.
A number of nonoscillatory results for the nonlinear equation (5.1.6),
require some sort of restriction on the growth of the function q(t),
typically that q(t) be nonincreasing. Other results demand the weaker
condition
'"Y
i- 1
dq+(s)
00
()
q S
to
< 00.
(5.4.16)
One such result is the following.

Theorem 5.4.2. Let condition (5.4.16) hold. Then equation (5.1.6) is
nonoscillatory provided anyone of the following is satisfied.
JOO sq(s)ds < 00,

lim qh- 1 )/2(t)
t-+oo
[00 s/'q(s)ds =
t
lim qh- 1)/(2 b +1(t)
t-+oo
00
0,
qh+l)h(s)ds
= 0,
lim t 2 q( t) = 0,
t-+oo
(Is)
lim t[OO q(s)ds=O.
t-+oo
The equation
x"(t)
+ t 2 (lnt)/' Ix(tW
sgn x(t) = 0,
< '"Y < 1
(5.4.17)
where t > to = 1 has a nonoscillatory solution x(t) = In t. Clearly,

conditions (1 2 ) - (15) are satisfied, but (1 1 ) fails. Motivated by this example,
Oscillation theory for sublineaJ" differential equations

we expect equivalence between (Ii)' i
the following lemma.
437
2,3,4,5. To prove this we need
Lemma 5.4.1. If q(t) E C([to, 00), R+) and is of bounded variation on

[to, 00), then it admits the decomposition
(5.4.18)
where ql(t) is a positive nonincreasing function, and q2(t) is a positive
nondecreasing function. Furthermore, if q(t) satisfies (5.4.16), then 0 <
kl < q2(t) ::::: k2 < 00 for all t, where kl and k2 are fixed positive
constants.
Proof. This follows from the identity
q(t) = q(to) exp (
-1: d~~;)S)) (1: d~;;;))
Here, q2(t) = exp Utto dq+(s)/q(s))

(5.4.16) holds.
Remark 5.4.1.
exp
tends to a finite limit as t ~ 00, if
Suppose condition (5.4.16) holds. Using the fact that
q2(t) is bounded from below and above by positive constants, we find that
q(t) satisfies anyone of the conditions (Ii)' 'i = 2,3,4,5 if and only if
ql (t) satisfies the same condition. Hence, in the proofs of some of the
results of this section, we can assume without loss of generality that q(t)
itself is nonincreasing.
Now, we are in the position to prove the following result.

Theorem 5.4.3. Under the growth assumption (5.4.16) (or more generally
(5.4.18)) on the function q(t), the following chain of implications holds
(5.4.19)
Proof. We shall prove that (14) <=} (15), (12) <=} (15),
(13)
(14). The case (II)
(14) is left to the reader.
'*
'*
Case 1. (14) <=} (15). The implication (14)

an application of L'Hopital's rule, i.e.,
o=
lim
t--+oo
'* (15)
(15)
'*
(13) and
follows trivially from
ft q(s)ds = lim -q(t) = lim t 2q(t).

OO
l/t
t--+oo -1/t2
t--+oo
Now, suppose that (15) holds, but (14) does not hold. Then there exists a
sequence {t n }, limn--+oo tn = 00 such that
t;q(tn) 2: 8 > 0 for some constant 8.
(5.4.20)
Chapter 5
438
By (15), if tn is large enough, then
(5.4.21)
On the other hand, since (5.4.20) holds and q(t) is nonincreasing q(t) 2:
for t E [t n /2, tn]. Integration of this inequality over [t n /2, t n ] yields
(tn/2) It:~/2 q( s )ds 2: J / 4, which contradicts (5.4.21).
Jt:;;2
Case 2. (b) <=> (15). We shall first show that (b)

f > 0, there exists a tl 2: to such that
=?
(15). Given any
(5.4.22)
Letting u(t) =
It" s'Yq(s)ds,
t'Y :S -
we obtain from (5.4.22),
f2/Cl-'Y)U-2/Cl-'Y) (t)u'(t),
Integrating this inequality from
tl
2:
tl.
to t 2: t l , we find
(5.4.23)
and hence
(5.4.24)
Now, an integration by parts from t to T 2: t 2: tl gives
(5.4.25)
It
The fact that

s'Yq(s)ds :S It" s'Yq(s)ds < 00, t 2: tl together with
(5.4.25) shows that oo q(s)ds converges. Using (5.4.24) and letting T-t
00 in (5.4.25), we obtain
00
Since
It
q(s)ds
= 0
(f
2/ C
1+'Y)C l )
as
-t 00.
> 0 is arbitrary, (5.4.26) establishes the validity of (15).
(5.4.26)
439
Now suppose that (Is) holds. Then for any E> 0 there exists t1 2': to
such that f tOO q(s)ds ::; Elt for all t 2': t 1 . An integration by parts from
t to T 2': t 2': t1 yields the following identity
lT
s'Yq(s)ds
t'Y
lT
The fact that
lT
q(s)ds +,
00
q(s)ds ::;
lT (ITq(~)d~)
s'Y- 1
q(s)ds =
(~)
as
t -+
ds.
(5.4.27)
00
implies that the right-hand side of (5.4.27), and hence the left-hand side,
converges as T -+ 00. Thus,
s'Yq(s)ds < 00. Furthermore, for t 2': tb
00
ftoo
s'Yq(s)ds ::; 1 ~ /"1- 1
et f-'Y ,
1
(5.4.28)
where E~-'Y = EI(l - ,). Now, for any t 2': t1, we need to consider the
two mutually exclusive subcases (i). q(t) 2': Edt 2 and (ii). q(t) < Edt2.
In the sub case (i), estimate (5.4.27) gives
In the sub case (ii), we set v = (Edq(t))1/2 > t, t 2': tl and observe
that q(t) is non-increasing. Then it follows from (5.4.28) that
qh- 1)/2(t)
Since
00
qh- 1)/2(t)
s'Yq(s)ds
::;
qh- 1)/2(t)
<
E1
h+1)/2
,+1
[l s'Yq(s)ds + 1 S'Yq(S)dS]
V
[1:
00
s'Yq(s)ds +
+ E~1-'Y)/2 =
(~) 1-"1]
O( y'E) as t -+
00.
> 0 is arbitrary, this establishes (h).
Case 3. (Is) =? (h). We consider the identity for 0 < , < J.L < 1,
lT
Since
sl-'q(s)ds = tl-'
lT
q(s)ds + J.L
ftoo q(s)ds = o(l/t)
as t -+
00
lT lT q(~)d~ds,
SI-'-l
(5.4.29)
00,
sl-'q(s)ds
T 2': t 2': to.
o(tl-'-l).
(5.4.30)
Chapter 5
440
By applying Holder's inequality, we have
r=
.Jt
ql/h+l)(s)ds:::
(1
00
sf-Lq(s)ds
1/h+ 1) (
Using (5.4.30) in (5.4.31), we find as t -+
00
r=
.Jt
s-f-Lhds
)1'/h+ 1)
(5.4.31 )
00,
ql/h+l)(8)ds ::: a (t(f-L-1)/h+ 1))
[(11 =I) th-~')hr/h+l)

=
a (th-1)/h+ l )) .
(5.4.32)
Now, for any

we have
E2
> 0 we can fix tl so that from (5.4.32) for all t 2: t l )
(5.4.33)
For any given t 2: to, we again consider two mutually exclusive sub cases
(i). q(t):;. C2/t2 and (ii). q(t) < Edt2. In the sub case (i), from (5.4.33),
we have the following estimate
(5.4.34)
In the subcase (ii), we set v
creasing, we have
(E2/q(t))1/2
> t. Since q(t) is nonin-
(5.4.35)
On the other hand, using estimate (5.4.33), we find
l/h+1) h-l)/(2h+1))
E2
E2
1/2
E2 .
(5.4.36)
Combining (5.4.35) and (5.4.36), we obtain
qh- 1)/(2(J+l))(t)
1=
ql/(1+,) (s)ds :::;
2E~/2.
(5.4.37)
Since E2 > 0 is arbitrary, (5.4.34) and (5.4.37) establish the desired assertion (13).
Case 4. (13)
(1+ 1)/(1- 1 )
=}
>
(14). Let
1. Choose
0:
Ea
= 2/(1 - I) > 1 and 1//3 = 0: - 1 =

< 4- a (0:-1). By (h) there exists tl 2: to

such that for all t 2': t 1 ,
q(r-l)/(2(r+l))(t)
100
ql/(r+ll(s)ds
441
<
E.
(5.4.38)
Denote f(t) = q/3/ 2 (t), t 2': t1, then fO(t) = ql/{l+'Y)(t). Suppose (14)
fails. Then there exist constants c and 6 > 0 such that c 2': 2t1 and
q(c) 2': 6/c 2 , or
(5.4.39)
where
61
from which it follows with d
= 61/(r+1) >
(5.4.40)
= 3c/4 that
(5.4.41)
Notice although the choice of c depends on E, 6 and 61 are independent

of E. We now rewrite (5.4.38) in terms of f(t) as
f(t) >
~ 100 r(s)ds
>
[i
r(s)ds
r(S)ds]
for tl :S c/2 :S t :S 3c/4. Using (5.4.41) in the above estimate, we find

(5.4.42)
Now, we define g(y)
becomes
= f(t) = f(d - V), so that for O:S y :S c/4, (5.4.42)
g(y) >
~61C-!3 + ~ fY gG(u)du.
4E
Jo
(5.4.43)
Comparing g(y) with its minorant function G(y) defined by
G(y)
~61C-i3 + ~ fY G""(u)du.
4E
Jo
(5.4.44)
we find as an application of a standard result in the theory of integral

inequalities that g(y) 2': G(y) for O:S y :S c/4. The function G(y) can
be found explicitly by solving (5.4.44) as
(5.4.45)
where
442
Chapter 5
From (5.4.45) it follows that G(y) ~

Thus, unless
00
_E (4E) 1/,13
a-I
bl
as y ~ (E/(a - 1(4E/b)I/13 c.
1
(5.4.46)
> 4'
G(x) will blow up on the finite interval [0, c/4]' and so will g(y). But,
this contradicts the fact that f(t) is defined on [c/2,3c/4J. Recall that
bl is determined only from the fact that (14) fails and is independent of
the choice of E. Hence, (5.4.46) must hold for all E > 0, an obvious
impossibility. This completes the proof of (13) =} (14) and also that of
Theorem 5.4.3.
Proof of Theorem 5.4.2 (1 2 ). Let x(t) be an oscillatory solution of

equation (5.1.6) and {tn} be the sequence of consecutive zeros of x(t).
We consider the energy function
E(x(t
[X'(t)J2
q(t)
= - - + --xl'+! (t),
'Y
+1
t::::: to.
Clearly, for the solution x(t) of (5.1.6), we have

d
dt E(x(t
'(~
(5.4.47)
!2(t) [x'(tW ::::: 0 for t::::: tf)
since q'(t) ::; 0 for t::::: to. From (5.4.47) it follows that {X'(tn)/ql/2(t n )}
forms a nondecreasing sequence, and in particular
where Co = x'(tO)/ql/2(tO). We let n to be sufficiently large, tn = r,

tn+l = rl so that
Without loss of generality we assume that x'(r)::::: 0 so that x(t) > 0 in
[r,rlJ.
Since x(t) is concave in [r,rl], x'(t)::; x'(r) for t E [r,rlJ. Thus, we

have x( t) ::; (t-r)x' (r) ::; tx' (r), t E [r, rlJ. Using this inequality in equation (5.1.6), we find -x"(t)::; t'Yq(t)[x'(r)p, t E [r,rd. Integration of this
inequality from r to t E [r, rlJ yields x'(r) -x'(t) ::; [x'(r)p
sl'q(s)ds.
In particular, when t = rl, we obtain
J:
443

(5.4.50)
Using (5.4.48) in inequality (5.4.50), we get c6-"Y q(1-"Y)/2 (T) ::;
fT= s"Y q( s )ds,
In case condition (5.1.16) does not hold, we have the following extension
of criterion (b).
Theorem 5.4.4.
Let q1 (t)
ftlo dq- (s )jq(s)). If
exp ( -
lim q;"Y- 1 )/2(t)

l-t=
1=
s"Yq(s)ds
0,
(5.4.51)

Proof. Let x(t) be an oscillatory solution of equation (5.1.6) and let
{t n } be the sequence of consecutive zeros of x( t). Consider the energy
function
E(x(t)) = [x'(tJF
q(t)
+ _2_X"Y+1(t)
,),+1
'
so that
::::: _ dq+(t) E(x(t)).

q(t)
(1
E(x(t)) ::::: E(x(to)) exp -
t
to
dq+(s))
(5.4.52)
where q2(t) is as in Lemma 5.4.1.

Let n be sufficiently large, tn = T, tn+1 = T1 so that
(5.4.53)
Without loss of generality we can assume that x' (T) ::::: 0 so that x( t) > 0,
t E [T,T1]. Now, in view of (5.4.52), we have [x'(t)j2jq(t) = E(x(r)) :::::
E(x(tO))jq2(T). From this inequality and Lemma 5.4.1, we find
(5.4.54)
Chapter 5
444
Proceeding as in Theorem 5.4.2 (1 2 ), we obtain (5.4.50). Using (5.4.54)

in (5.4.50), we get [E(x(to1(1-I')/2q~1-I')/2(T)::; fT(X) sl'q(s)ds, which
contradicts (5.4.53). This completes the proof.
5.5. Oscillation Criteria for Certain

In this section we shall consider the nonlinear differential equation
(a(t)lxl(t)I"'-lXI(t)1
+ q(t)f(x(t =
0,
(5.5.1)
where
(i) 0: is a positive constant,
(ii) aCt) E C([to ,oo),1R+), q(t) E G([to, 00), 1R),
(iii) f E G(1R, 1R), xf(x) > 0 and f'(x)::::: 0 for x
-I- O.
In what follows we shall denote by
1](t)
ltD
a- 1/", (s)ds
and assume that
lim 1](t) =
t--'>(X)
00.
(5.5.2)
We shall relate the oscillation results for (5.5.1) to those of half-linear

differential equations discussed in Chapter 3. We shall also present some
sufficient conditions as well as necessary and sufficient conditions for the
oscillation of (5.5.1) and/or related equations.
We begin with a special case of (5.5.1), namely, the equation
(a(t)lxl(t)I"'-lXI(t)1
+ q(t)lx(t)I13-1 X(t)
0,
(5.5.3)
where (J is a positive constant.

Theorem 5.5.1. Let (J
< 0: and condition (5.5.2) hold, and
liminf
t--'>co
ltD
q(s)ds > - 00.
(5.5.4)
If for every constant k > 0, the half-linear differential equation
(a(t)1]"'(t)l yl(t)I",-lyl(t1 + kq(t)ly(t)I"'-ly(t) = 0
(5.5.5)

x(t) > 0 for t::::: to. Define wet) = a(t)lxl (t)I"'-lx l (t)/x 13 (t) for t::::: to.
Proof.
445
Then for t;:::: to, we have
w'(t)
Ix'(t)I<>+1
-q(t) - /3a(t) x,B+1(t)
(5.5.6)
-q(t) - /3a- 1 /a(t)lw(t)l(a+1)/a (x(,B-a)/a(t))

and hence
w(t) = w(to) -
to
q(s)ds - /3
it
to
a(s)
Ix'(s)la+1
,B+1() ds.
X
(5.5.7)
It:
We distinguish two mutually exclusive cases where the integral I(t) =

a(s)lx'(s)la+l /x,B+1(s)ds remains finite or tends to infinity as t -+ 00,
and arrive in each case at a contradiction.
Case 1. I( (0) < 00. In this case there exists a positive constant Cl such
that I(t):S C1 for t;:::: to. For t;:::: to, we use Holder's inequality, to
obtain
x 1--r(t) - x 1--r(to)
it I
L
= (-y - 1)
:S (-y - 1)[I(t)]1/(a+1)
x'(s) IdS
x-r (s)
] a/(a+1)
a- 1/a(s)ds
to
where 'Y = (/3 + 1)/(0: + 1) < 1 and C2 = (-y -l)c~/(<>+l). By condition

(5.5.2) there exist a constant C3 > 0 and aT;:::: to such that x 1--r(t):S
C37J a /(a+1) (t), or x(t):S (C37J a /(<>+1)(t))1/(1--r) for t;:::: T and
x(,B-a)/<>(t) > ~
- 7J(t)
where
w'(t)
C4
for
t> T
-
(5.5.8)
(C3)(,B-a)/(a(1--r)). Using (5.5.8) in equation (5.5.6), we get
+ q(t) + (C~)
0:
(a(t)~a(t)) 1/<> Iw(t)l(a+1)/<>
:S 0
for
t;:::: T.
Now, applying Lemma 4.4.4 we find that (5.5.5) is nonoscillatory, which is

a contradiction.
Case 2. I(oo) = 00. Following exactly as in Theorem 4.4.3 Case 2,
we arrive at the contradiction limt->oo x(t) = -00. This completes the
proof.
To extend Theorem 5.5.1 to general equations of type (5.5.1), we need

the following conditions
(5.5.9)
446
Chapter 5
and
min { inf
x>o f
1'(x) lX
1
inf 1'(x)
~
fl/a( ) du, x<o
fY(x)
(x) +0
u
Q
jX
-0
p/a(u)
du }
> O.
(5.5.10)
Remark 5.5.1. Conditions (5.5.9) and (5.5.10) when 0: = 1 reduce to
(5.1.5) and (5.2.86). Also, we can replace condition (5.5.10) by
l' (x ) )
( f(a-l)/a(x)
1
---,--,,------,-...,...du
> -1 > 0 for all
p/a(u)
c
x,
(5.5.11)
where c is a constant.
N ow we sha.ll prove the following theorem.
Theorem 5.5.2. Let conditions (5.5.2), (5.5.4), (5.5.9) and (5.5.10) hold.
If for every constant k > 0, the half-linear differential equation (5.5.5) is
oscillatory, then (5.5.1) is oscillatory.

x(t) > 0 for t 2: to. Define wet) = a(t)lx'(t)la- 1 x'(t)/ f(x(t, t 2: to.
Then for t 2: to, we have
Proof.
Ix'(t)laH ,
w'(t) = - q(t) - aCt) j2(x(t f (x(t),
or
W '(t)
q(t) - a-1/a(t)lu'(t)l(aH)/a (
1'(x(t
)
fCa-l)/a(x(t
'
(5512)
..
and hence
rt
rt
Ix'(s)laH,
wet) = w(to) - lto q(s)ds - lto a(s) j2(x(s) f (x(sds.
Now we need to consider two cases, when
I(to,t)
Ix'(s)la+l,
lto a(s) j2(x(s)) f (x(sds
remains finite or tends to infinity as t ---+

Case 1. I(to, 00) <
such that
00.
00.
In view of (5.5.10), there exist a constant c> 0
1'(x(t
) lxCt)
1
( JCa-l)/a(x(t
+0 p/a(u) du > c for t 2: to
(5.5.13)
447
We will show that

lim v(t) = 0
Hoo
where v(t) = J:~t) du/ jI/rY.(u). For this, let

Then, we can choose a t1 ::::: to such that
I(t1'00):S:
By Holder's inequality, for every
v(t)-v(t 1) <_ lIt

t,
(5.5.14)
ry(t)
> 0 be an arbitrary number.
Ec)a+1
( 2"
(5.5.15)
t::::: t1, we find
x'(s) dsl < [I(t t)]l/(a+1)[q(t t)]a/(a+1)

Jl/a(x(s)
1,
J
1,
,
where
_
:l(t1' t) -
It -l/a
(f(a-1)/a x (s)
t, a
(s)
f'(x(s)
l/a
ds,
and consequently, by using (5.5.15), we get
(5.5.16)
If :l(t1, 00) < 00, (5.5.16) ensures that v(t) is bounded and hence (5.5.14)
holds. So, we restrict our attention to the case when :l(t1'00) = 00. Let
t2 > t1 be such that V(t1):S: (cE/2)[.J(t1' t)]a/(a+1) for t::::: t2' Then
(5.5.17)
and hence, from (5.5.13), we obtain
v(t)
:s:
c1/(a+1)E
[1.
a- 1/a(s)v1/a(s)ds
a/(a+1)
or
Integrating inequality (5.5.18) from t2 to t with
0:0 : 1
-
[It
t, a- 1/a(s)v1/a(s)ds ]
:s:
0+'
0:0 : 1
-
c1/(a+1)E (t a- 1/a(s)ds
it2
[l t2
t,
:s:
t::::: t2,
we get
a- 1/rY.(s)v1/a(s)ds ]
c1/(a+1)Ery(t).
0+'
Chapter 5
448
Now, there exists a t3 2': t2 such that
[
11 a-1/a(s)v1/a(s)ds
t
] a/(aH)
< CIE"I(t) for all
where Cl = 2(a/(a + 1))c1/(,,+1). Thus, from (5.5.17) it follows that

v(t) < CCIE2'1}(t) for all t 2': t3. As E> 0 is arbitrary, the proof of (5.5.14)
is complete.
Now from (5.5.13) and (5.5.14) there exist a T 2': t3 and a constant
C2 > 0 such that
f'(x(t))
2': C2
f(a-l)/a(x(t))
'I}(t)
for
2': T.
(5.5.19)
Using (5.5.19) in (5.5.12), we obtain
w'(t)
+ q(t) + c2(a(t)'I}"(t))-1/<>lw(t)l(a+l)/" :s:;
for all
t 2': T.
Finally, applying Lemma 4.4.4 we find that (5.5.3) is nonoscillatory, which

is a contradiction.
Case 2. I( to, (0) = 00. For this case the proof can be modelled on Theorem
4.4.3 case 2. We omit the details.
Remark 5.5.2. By Theorems 5.5.1 and 5.5.2 it is clear that the results
established in Chapter 3 for half-linear differential equations can be employed to obtain some interesting oscillation criteria for nonlinear equations
of types (5.5.1) and (5.5.3). The formulation of such results are left to the
reader.
Remark 5.5.3. From the results presented in earlier chpaters it is clear
that condition (5.5.4) in Theorems 5.5.1 and 5.5.2 can be weakened.
In the following results we shall present oscillation criteria for (3.13.76)
and (3.13.89) with g(t) = t, i.e., we shall consider the equations
(a(t)~(x'(t)))'
+ F(t, x(t)) =
(5.5.20)
0,
(5.5.21)
and
(a(t)(x'(t))a*)' +F(t,x(t))
where the functions a, F, ~ and the constant a

(3.13.76) and (3.13.89), and (x')"* = Ix'I" sgn x'.
are as in equations
Definition 5.5.1. We say that equation (5.5.20) is strongly sublinear if

there exists a constant c> 0 such that lyl-CIF(t,y)1 is nonincreasing in
Iyl for each fixed t, and
(5.5.22)
449
Equations (5.5.3) and (5.5.21) are strongly sublinear if j3 <

Q
respectively holds.
o< c <
and
In what follows with respect to equation (5.5.20) we shall assume that
1~ 1'I/J-1 (a~)) 1ds
00
for every constant
~ o.
(5.5.23)
Theorem 5.5.3. Let equation (5.5.20) be strongly sublinear. Suppose for

each fixed k ~ 0 and to ~ 0,
1UTI
-+0, k>0
\II,to(a, t)
0
=
t)
(5.5.24)
\II k,to (a,
uniformly on any interval of the form ttl, (0) for some t1 > to. Suppose
moreover that
(5.5.25)
00
IF(s, c\llk(a, s))lds
(5.5.26)
00
to
for all constants k

Section 3.13.4.
0 and c > 0, where the function \II is defined in
Proof. The 'only if' part is a consequence of Theorem 3.13.11. To prove

the 'if' part, let x(t) be a positive solution of equation (5.5.20) for t ~ to.
First, we note that x'(t) > 0 for t ~ to, and
x'(t)
~ ~r1
(att))

a(t)'I/J(x'(t)) , we find
~r1(a(t)'I/J(x'(t)))
to
to t
for
G(u) =
~ to
(5.5.27)
and using the decreasing nature of
x(t) - x(to) > 'I/J-1 (a(t)'I/J(x'(t)))\II1,to (a, t),

Now define
uo
t ~ to.
dv
['I/J-1(uW'
(5.5.28)
(5.5.29)
where c> 0 is the strong sublinearity constant and uo = a(to)'I/J(x'(to)).

Then from (5.5.28) and (5.5.29) for t ~ to, we get
[G( a( t)'Ij!(x' (t))) l'
>
F(t,x(t))
['I/J-1 (a(t)'I/J(x'(t)) W
['l1 1,to (a, tW(x( t))-C F(t, x(t)).
(5.5.30)
Chapter 5
450
From the strong sublinearity and the inequality x(t) :::; COWk,to(a, t), t 2: to
where Co > 0 is a constant, it follows that
(5.5.31)
for t > to. Substituting (5.5.31) into (5.5.30) and using the inequality
Wl,to(a, t) >
Wk.to(a,t) -
1p-l
(.!.)
t > to
k'
which follows from (5.5.25), we obtain by integration over [h, t 2 ], tl > to

that
O 1p-l
(~) lt2 F(s, COWk,to(a, sds
1~1 [1P-~~'VW'
:::;
where Ui = a~i)w(x'(ti' i = 1,2. Letting t2 -t 00 and using (5.5.22),

we obtain
F(s, COWk,to(a, sds < 00, which contradicts condition
hI
The following result is a. va.riant of Theorem 5.5.3.

Theorem 5.5.4. Assume that there exist functions q(t) E CUto, 00), JR+)
and g(x) E C(JR, JR) such that g(x) is increasing,
IF(t, y)1 2: q(t)g(IYI)
for
(t, y)
JR+
(5.5.32)
1R.
Moreover, suppose that conditions (5.5.24) and (5.5.25) hold,
g(xy) 2: g(x)g(y)
and
du
i+o go Iw-1(u)1
<
00
and
x, y > 0
for any
-0
go
1 dU1 ( )1 <
w-
(5.5.33)
00.
(5.5.34)
00
q(S)g(Wk(a, sds
(5.5.35)
00
to
for every constant k
-I- O.
Proof. The 'only if' part is a consequence of Theorem 3.13.11. To prove

the 'if' part, let x(t) be a positive solution of equation (5.5.20) for t 2: to.
As in Theorem 5.5.3, (5.5.28) holds, and so (5.5.33) implies
g(x(t
2: g(w-1(a(t)w(x'(t)g(wl,to(a, t,
t 2: to
(5.5.36)
Oscillation theory for sublinea,r differential equations
Define
dv
10
H(u) =
451
gO'lj;-l(V)'
(5.5.37)
u> 0
where Uo = a(to)'lj;(x'(to)) > O. Then by (5.5.36) and (5.5.37) for t > to,
we have
F(t, x(t))
g( ~/r 1 (a( t )~j;( x' (t)))
q(t)g(x(t))
;::: q(t)g(\}il,to(a, t)).
>
g( ~p-l (a(t)~j;(x'(t)))
(5.5.38)
Integrating (5.5.38) from t1 > to to t2 > tI, we find
[H (a( t )~j;( x' (t)))]'
t2 q(S)g(\}il,to(a, s))ds
t,
where Ui
:s: 111
U2
= a(ti)'lj;(X'(ti)),
It"; q(S)g(\}il,to(a, s))ds < 00,
0:~l( )'
~
= 1,2 which in view of (5.5.34) implies

J'X)
ExaIllple 5.5.1. Consider (5.5.3) and assume that

a-1/0I(s)ds = 00.
Clearly, from Theorem 5.5.3 if IX > (3, a necessary and sufficient condition
for equation (5.5.3) to be oscillatory is I:: s!3q(s)ds = 00.
Next, with respect to equation (5.5.21) we shall assume that
7l'(t)
00
a- 1 /0I(s)ds
and
7l'(to) <
00.
(5.5.39)
Theorem 5.5.5. Let equation (5.5.21) be strongly sublinear. Then it is

oscillatory if and only if
lOO (ats)
1:
IF(u,k)ldU) 1/01 ds
00
(5.5.40)
for every constant k =1= O.

Proof. It suffices to prove the 'if' part, since the 'only if' part follows from
Theorem 3.13.16. Suppose condition (5.5.40) holds and equation (5.5.21)
has an eventually positive solution x(t), t :2: to. We consider the two cases:
Case I. If x'(t) > 0 for t:2: to, then we have Itr:;: F(s, cl)ds < 00,
where Cl > 0 is a constant such that x(t):2: Cl for t;::: to. This together
with the condition (5.5.39) shows that
1 (1
00
to
a(s) Jto F(u, cl)du
)1/0.
ds:S: 7l'(to)
( JtorOO F(s, cr)ds ) 1/01
<
00,
Chapter 5
452
Case II. If x'(t) < 0 for t ~ tl ~ to, then an integration of (5.5.21) gives
F(s, x(sds ::; a(t)( -x'(t'"
for
t,
~ tI ,
or equivalently,
-x'(t)
Since, x(t) ::;
implies
It
)
( aCt) l F(s,x(sds
~
C2,
1/",
t ~ t l , for some constant
C2
(5.5.41 )
> 0, the strong sublinearity

(5.5.42)
for some constant c < a (see Definition 5.5.1). Combining (5.5.41) with
(5.5.42) and using the decreasing property of x(t), we find
-x'(t)
> c;-C/"'a-I/"'(t)
(t)'
= _
>
xC(s)F(s, C2 )dS) 1/",
c;-C/"'a-I/<>(t)xC/<>(t)
Differentiating the function
(-x a;:c
(1:
(a:
(1:
F(s,C 2)dS) I/n,
~ tl.
(5.5.43)
- x(<>-c)/<>(t) and using (5.5.43), we obtain
c) x-c/<>(t)x'(t)
(a: C) c;-c/ax-8/", (t)a-I/a (t)xc/<> (t) (ltF(S, C2)dS) 1/<>

(a: C) c;-C/<>a-1/<>(t) (1: F(s, C2)dS) 1/<>, t ~ t1
from which we have by integration
(a: C) c;-c/<> 1:
1:
lOO (a~) 1:
2
(ats)
for any t2 > t1. Letting
F(u, c2)du
t2 -4 00
y/a
ds ::; x(a-c)/a(t 1) (5.5.44)
in (5.5.44), we conclude that
F(U,C2)duy/a ds <
00,
which again contradicts condition (5.5.40). This completes the proof.
Example 5.5.2. Consider (5.5.3) with aCt) = eAt and q(t) = el-'t for
t > 0, where >. and JL are real constants and >. > O. Let Q: > (3. Then
by Theorem 5.5.5, equation (5.5.3) is oscillatory if and only if JL
>..
453

1. Theorems 5.1.1 and 5.1.2 extend sublinear oscillation criteria of
Kamenev [28] and Kusano et. a1. [30]. Theorem 5.1.3 is due to Grace
and Lalli [13] and it includes theorem 8 of Graef et. a1. [26]. Theorems
5.1.4 and 5.1.5 extend the corresponding results of Wong [42]. Theorem
5.1.6 is taken from Butler [5], whereas Theorem 5.1.7 is based on the work
of Kwong and Wong [33] and it generalizes Belohorec criterion [3].
2. Condition (F) and Examples 5.2.1 - 5.2.3 are due to Philos [36].
Theorems 5.2.1 - 5.2.3 are extracted from Grace and Lalli [20]. Corollaries
5.2.1 and 5.2.2 are the same as theorems 2 and 3 of Philos [37]. Also
Corollary 5.2.3 is the same as theorem 1 of Philos [40]. Theorems 5.2.4 and
5.2.5 are new. Corollaries 5.2.4 and 5.2.7 are the same as theorems I' and
2' of Philos [38]. Corollary 5.2.8 is the same as theorem 1 of Kwong and
Wong [31]. Theorems 5.2.6, 5.2.7 and 5.2.9 are new while Corollary 5.2.10
is the same as theorem 2 of Philos [40]. Theorem 5.2.8 is due to Philos [39].
Theorem 5.2.10 is based on the work of Wong [48]. Theorem 5.2.11 is new
while Corollary 5.2.13 is due to Philos and Pumaras [41]. Theorems 5.2.12
and 5.2.13 are new while Corollary 5.2.16 is borrowed from Wong [48].
3. Theorems 5.3.1 - 5.3.3 are taken from Agarwal et. a1. [1]. These
results extend as well as improve theorems 4 - 6 of Kwong and Wong [32].
4. The results of Section 5.4 are taken from Agarwal et. a1. [2], Gollwitzer [7], Kwong and Wong [31,34] and Wong [43,49].
5. Theorems 5.5.1 and 5.5.2 are new. Theorems 5.5.3 and 5.5.4 are
taken from Elbert and Kusano [6] whereas Theorem 5.5.5 is due to Kusano
et. a1. [29].
6. Several other related oscillation results for sublinear differential equations are available in Butler [4], Grace [8-12], Grace and Lalli [14-19,21-24]'
Grace, Lalli and Yeh [25], Heidel [27], Onose [35], and Wong [44-47].
7. It will be interesting to discuss oscillation criteria for forced and
damped sublinear differential equations similar to those dealt with in Sections 4.5 and 4.6.
5.7. References
1. R.P. Agarwal, S.R. Grace and D. O'Regan, Linearization of second
order sublinear oscillation theorems, to appear.

2. R.P. Agarwal, S.R. Grace and D. O'Regan, Oscillation Theory for
3. S. Belohorec, Oscillatory solutions of certain nonlinear differential equations of second order, Mat. Fyz. Casopis Solven. Akad. Vied. 11(1961),
454
Chapter 5
25(}-255.
4. G.J. Butler, Integral averages and the oscillation of second order ordinary
differential equations, SIAM J. Math. Anal. 11(1980), 19(}-200.
5. G.J. Butler, An integral criterion for the oscillation of a second order
sublinear ordinary differential equations, Indian J. Math. 24(1982), 1-7.
6. A Elbert and T. Kusano, Oscillation and nonoscillation theorems for a
class of second order quasilinear differential equations, Acta. lvlath. Hungar. 56(1990), 325-336.
7. H.E. Gollwitzer, Nonoscillation theorems for a nonlinear differential equ-
ation, Proc. Amer. Math. Soc. 26(1970), 78-84.
8. S.R. Grace, Oscillation theorems for second order nonlinear differential
equations with damping, Math. Nachr. 141(1989), 117-127.
second order, J. Math. Anal. Appl. 171(1992), 22(}-241.
10. S.R. Grace, Oscillation theorems for damped functional differential equations, Funkcial. Ekvac. 35(1992), 261-278.
11. S.R. Grace, Oscillation of nonlinear differential equations of second order,
Publ. Math. Debrecen 40(1992), 143-153.
12. S.R. Grace, On the oscillatory behavior of solutions of second order nonlinear differential equations, Publ. Math. Debrecen 43(1993), 351-357.
13. S.R. Grace and B.S. Lalli, Oscillation theorems for certain second order
205-214.
14. S.R. Grace and B.S. Lalli, Oscillation of solutions of damped nonlinear
second order functional differential equations, Bull. Inst. lvlath. Acad.
Sinica 12(1984), 5-9.
15. S.R. Grace and B.S. Lalli, Oscillation theorems for a second order nonlinear ordinary differential equation with damping term, Comment. Math.
Univ. Caralin. 27(1986), 449-453.
16. S.R. Grace and B.S. Lalli, Oscillatory behavior of solutions of second order differential equations with alternating coefficients, Math. Nachr.
127(1986), 165-175.
17. S.R. Grace and B.S. Lalli, An oscillation criterion for certain second order strongly sublinear differential equations, J. Math. Anal. Appl.
123(1987), 584-588.
18. S.R. Grace and B.S. Lalli, An oscillation criterion for second order
sublinear ordinary differential equations with damping term, Bull. Polish.
Acad. Sci. Math. 35(1987), 181-184.
19. S.R. Grace and B.S. Lalli, Oscillation theorems for second order nonlinear differential equations, J. Math. Anal. Appl. 124(1987), 213-224.
20. S.R. Grace and B.S. Lalli, On the second order nonlinear oscillations,
Bull. Inst. Math. Acad. Sinica 15(1987), 297-309.
second order nonlinear differential equations, Ann. Mat. Pura Appl.
151(1988), 149--159.
Oscilla.tion theory for sublinear diHerential equations
455
22. S.R. Grace and B.S. Lalli, Oscillations in second order differential equations with alternating coefficients, Period. Math. Hungar. 19(1988), 6978.
23. S.R. Grace and B.S. Lalli, Oscillation theorems for nonlinear second
order differential equations with a damping term, Comment. lvIath. Univ.
Carolin. 30(1989), 691-697.
24. S.R. Grace and B.S. Lalli, Integral averaging techniques for the oscillation of second order nonlinear differential equations, J. Math. Anal. Appl.
149(1990), 277-311.
SIAM J. Math. Anal. 15(1984), 1082-1093.
375-390.
27. J.W. Heidel, A nonoscillation theorem for a nonlinear second order differential equation, Proc. Amer. Math. Soc. 22(1969), 485-488.
28. LV. Kamenev, Oscillation of solutions of second order nonlinear equations with sign variable coefficients, Differencial'nye Uravnenija 6(1970),
1718-1721.
partial differential equations, Japan J. IvIath. 19(1993),131-147.
30. T. Kusano, H. Onose and H. Tobe, On the oscillation of second order
nonlinear ordinary differential equations, Hiroshima Math. J. 4(1974), 491499.
31. M.K. Kwong and J.S.W. Wong, On the oscillation and nonoscillation
of second order sublinear equations, Proc. Amer. Math. Soc. 85(1982),
547-551.
32. M.K. Kwong and J.S.W. Wong, Linearization of second order nonlinear oscillation theorems, 'nans. Amer. Math. Soc. 279(1983), 705-722.
33. M.K. Kwong and J.S.W. Wong, On an oscillation theorem of Belahorec, SIAM J. Math. Anal. 14(1983), 474-476.
34. M.K. Kwong and J.S.W. Wong, Nonoscillation theorems for a second
order sublinear ordinary differential equation, Proc. Amer. Math. Soc.
87(1983), 467-474.
35. H. Onose, On Butler's conjecture for oscillation of an ordinary differential
equation, Quart. J. Math. 34(1983), 235-239.
36. Ch.G. Philos, Oscillation of sublinear differential equations of second
order, Nonlinear Analysis 7(1983), 1071-1080.
37. Ch.G. Philos, On second order sublinear oscillation, Aequations Math.
27(1984), 242-254.
38. Ch.G. Philos, Integral averaging techniques for the oscillation of second
order sublinear ordinary differential equations, J. Austral. Math. Soc. Ser.
A 40(1986),111-130.
456
Chapter 5
39. Ch.G. Philos, On the oscillation of second order sublinear ordinary differential equations with alternating coefficients, Math. Nachr. 146(1990),
105-116.
40. Ch.G. Philos, Integral averages and oscillation of second order sublinear
differential equations, Differential and Integral Equations 4( 1991), 205-213.
41. Ch.G. Philos and I.K. Purnaras, On the oscillation of second order
nonlinear differential equations, Arch. Math. 59(1992), 260-271.
42. J .S.W. Wong, Oscillation theorems for second order nonlinear differential
equations, Bull. Inst. Math. Acad. Sinica 3(1975), 283-309.
43. J .S.W. Wong, Remarks on nonoscillation theorems for a second order
nonlinear differential equation, Proc. Amer. Math. Soc. 83(1981), 541546.
44. J .S.W. Wong, Oscillation theorems for second order nonlinear differential
equations, Proc. Amer. Math. Soc. 106(1989), 1069-1077.
45. J .S.W. Wong, A sublinear oscillation theorem, J. Math. Anal. Appl.
139(1989), 197-215.
46. J .S.W. Wong, An oscillation theorem for second order sublinear differential equations, Proc. Amer. Math. Soc. 110(1990), 633-637.
47. J .S.W. Wong, Oscillation criteria for second order nonlinear differential
equations with integrable coefficients, Proc. Amer. Math. Soc. 115(1992),
389-395.
48. J .S.W. Wong, An oscillation criterion for second order nonlinear differential equations with iterated integral averages, Differential and integral
Equations 6(1993), 83-91.
49. J.S.W. Wong, Nonoscillation theorems for second order nonlinear differential equations, Proc. Amer. Math. Soc. 127(1999), 1387-1395.
Chapter 6
Further Results on the
Oscillation of Differential
Equations
6.0. Introduction
In this chapter we shall discuss some techniques which are different
than those employed in the previous chapters to obtain oscillatory criteria
for differential equations. In Section 6.1 we shall present oscillation and
nonoscillation theorems for nonlinear second order differential equations by
using the method of Olech, Opial and Wazewski. Section 6.2 is concerned
with the oscillation of half-linear second order differential equations by employing the variational inequality given in Lemma 3.2.6. In Section 6.3 we
shall begin with some preliminaries of Liapunov functions, and then apply Liapunov second method to obtain criteria for the oscillation of second
order nonlinear equations.
6.1. Oscillation Criteria of

Olech-Opial-Wazewski Type
Consider the second order nonlinear differential equation
x"(t)
+ q(t)f(x(t))
(6.1.1)
and its prototype, the so called generalized Emden-Fowler equation
x"(t) + q(t)jx(t)j'Y sgn x(t)
0,
(6.1.2)
where I > 0 is a constant.

In what follows we shall assume that
(i) f E C2 (lR- {O},lR), xf(x) > 0 for x
(ii) q E C(lRo, lR).
i= 0
and f'(x) > 0 for x
i= 0,
Chapter 6
458
As a main result of this section we shall extend an oscillation result of

Olech, Opial and Wazewski [8]. They showed that equation (6.1.2) with
'Y = 1 is oscillatory if
limapprox
t -+ 00
fat q(s)ds = 00,
(6.1.3)
in
or
lim approxinf
t -+ 00
ft q(s)ds <
limapproxsup
t -+ 00
io
(6.1.4)
fat q(s)ds,
in
(see definitions in the next subsection).

As a consequence of conditions (6.1.3) and (6.1.4) it follows that if
t
io
lim approx sup

t -+ 00
q(s)ds
(6.1.5)
00,
then equation (6.1.2) with 'Y = 1 is oscillatory.
6.1.1. Some Definitions and Notation

For any set S
c ]Ro, we define the density function of
S by
ps(t) = iP,{S n [0, t]},
(6.1.6)
where p, denotes Lebesgue measure. We denote by g(x) the expression
g(x)
f"(x)f(x)
(f'(x))2'
Clearly, g(x) is continuous for x

> 0 is a constant, then
=1=
g(x)
'Y
o.
=1=
o.
(6.1. 7)
We note that if f(x)

'Y - 1
'Y
= Ixll' sgn x,
(6.1.8)
With respect to the function g(x), we will assume that the following
condition holds: There exist numbers M > 0 and m < 1 such that
g(x) ::; m < 1 and
Ig(x)1 ::; M
for all
=1=
o.
(6.1.9)
Now, we recall the following definitions: If h = h(t) is a real valued

function defined on [a,oo), a 2: 0 and if - 00 ::; , L ::; 00, then
we write limapproxsupHOOh(t) = L if p,{t: h(t) > Ld = 00 for all
L1 < Land p,{ t : h(t) > L 2} < 00 for all L2 > L. Similarly, we write
limapproxinfHooh(t) = if p,{t: h(t) < 1} < 00 for all 1 < and
p,{t: h(t) < 2} = 00 for all 2> . We write limapproxHooh(t) = L,
Further results on the oscillation of differential equations
459
-00 ::; L ::; 00 if limapproxsuPt-+ooh(t) = limapproxinft-+ooh(t) = L.

We note that limapproxt-+ooh(t) = 00 implies that there exists a set
S C lRo such that limt-+oo, tES h(t) = 00 and limsuPt-+oo ps(t) =
limt-+oops(t) = 1.
6.1.2. Oscillation Criteria

Our main result here is the following:
Let condition (6.1.9) hold and assume that there exists
a set S C lRo such that
Theorem 6.1.1.
lim sup t[ps(t) - fj(M, m)]

t-+oo
b(M, m)
00,
M2 -4m+4
(6.1.10)
(6.1.11)
and
lim
t-+CXJ, tES
q(s)ds
io(
00.
(6.1.12)

Proof. Suppose (6.1.1) is not oscillatory, and let x( t) be a nonoscillatory
solution on lRo, which we suppose satisfies x(t) > 0 on [to, (0), to 2: O.
Define w(t) = x'(t)/f(x(t)), t 2: to. Then for t 2: to, we have
w'(t)
q(t) - f'(x(t))w 2 (t),
(6.1.13)
which on integration over [to, t] gives
w(t)
w(to)
-it -it
q(s)ds
to
f'(x(s))w 2 (s)ds.
(6.1.14)
du
f(u)'
(6.1.15)
to
Define the function 0(x), x> 0 by
0(x)
ix
Since 0(x) is decreasing, 0- 1 exists. Further, since
it
to
w(s)ds =
where Co
it
to
x'(s)
f(x(s)) ds = -
r(t
ix(t)
o)
du
f(u) = - 0(x(t))
+ co,
0(x(t)),
(6.1.16)
0(x(to)), if we set
y(t)
Co
-it
to
w(s)ds
460
Chapter 6
then x(t)
(6.1.l4),
= n- 1(y(t)) and y'(t) = -w(t),
q(s)ds
y'(t)
-it
t 2
to and so we have from
f'(n- l (y(s))(y'(s))2ds - y'(to)
(6.1.17)
to
to
t 2 to
y'(t) - W(t) - y'(to),

where
W(t) =
.f:
f'(n-1(y(s))(y'(s))2ds,
2 to
(6.1.l8)
Now (6.1.12) implies that q(t) is not identically zero on any half-line
(tl, (0), and so y'(t) is not identically zero on any half-line (tIl (0)
for some t1 2 to. It follows that W (t) > 0 for all t sufficiently large.
Without loss of generality we may assume that W(t) > 0 for t 2 to. We
now let the function P(t) be defined by the equation
y'(t) = P(t)W(t)
for
> to.
(6.1.l9)
We then have
W'(t) = (y'(t))2 f'(n- 1(y(t))) = p2(t)W 2(t)f'(x(t))
for
t > to.
If we let V(t), U(t) be defined by the relations
V(t) = f'(x(t)) = f'(n- 1(y(t))), t 2 to
and
U(t) =
V(t)~(t)'
> to,
we find that V(t) > 0 and U(t) > 0 for t > to, and
V'(t)
j"(x(t))x'(t)
j"(x(t)) :t (n-1(y(t)))
(6.1.20)
- f"(x(t) )f(x(t) )y' (t)

- 1"(x(t))f(x(t))P(t)W(t)
for
> to
We also have
U'(t)
V'(t)
W'(t)
V2(t)W(t)
V(t)W2(t)
1" (x( t) )f(x( t) )P( t) W (t)
p 2(t) W 2(t)f' (x(t))
V2(t)W(t)
V(t)W2(t)
g(x(t))P(t) - p 2(t) for t > to,
(6.1.21)
where g(x) is defined by (6.1.7).

Now for any natural number k> 0 from (6.1.12) we have J~ q(s)ds 2
k for t E S n [tk' (0) if tk 2 to is sufficiently large. Define the set Sl by
Sl
{t > to: P(t) 21}.
(6.1.22)
461
TheIl from (6.1.17) and (6.1.19) for all sufficiently large k > 0, we have
(6.1.23)
Assume now that k is fixed so that tk > to and (6.1.23) holds, and let
5 = Sl n [tk' (0). It follows that
Ji(S
.
1lmsup
n [0, t])
t
1--'>00
Now on
5,
1.
Ji(S
Ilnsup
t--'>oo
n [tk' t])
t
Ji(5n[tk,t])
.
<::: 1mlSUp
t--'>oo
we have from (6.1.21) and (6.1.9) that
U'(t) = P(t)(g(x(t))-P(t)) <::: P(t)(rn-P(t)) <::: rn-I < 0, for t E 5
(6.1.21)
(since the function rnP - p 2 for P 2: 1 is maximized at P = 1). Further
on 5c = [tb cc)\5 (= the complement of 5 in [tk' (0)), P(t) < 1 so that
the function P(a - P) is maximized at P = a/2. Thus, from (6.1.21)
and (6.1.9), we obtain
1
U'(t) <::: ?(g(x(t)))2

- -(g(x(t)))2
= -(g(X(t)))2
<
_
4
4
If we set
5=
U(t)
n [tk' t], t > tk

U(tk) + (
}s(t)
-4]1;[2
for
t E
5c .
(6.1.25)
then we have
U'(s)ds + (
}se(t)
U'(s)ds,
t 2: tk.
Since U(t) > 0, t > to it follows that
1 set)
U'(s)ds
+ (
for some constant L>
}se(t)
U'(s)ds 2: - L > -
and all
(6.1.26)
00
t 2: tk. From (6.1.24), we have
U'(s)ds <::: (rn-1)Ji(5(t)) <::: t(rn-1)ps(t) for all large t (6.1.27)
1s(t)
and from (6.1.25), we obtain

for all large t.
(6.1.28)
Therefore, from (6.1.26) - (6.1.28), we get
t [(rn -l)Ps(t)
+ ~M2(1- Ps(t))] 2: -
L > -
00
for all large t,
Chapter 6
462
which implies
4L
M2]
t [PS(t) - M2 _ 4m + 4
< M2 -4m+4 <
contradicting condition (6.1.11). This proves the theorem.
00
Corollary 6.1.1. Let condition (6.1.9) hold and

limapprox
t -+ 00
t
io
q(s)ds =
(6.1.29)
00.

Proof.
If (6.1.29) holds, then condition (6.1.12) is satisfied for some

1Ro Since !5(M, m) < 1 in (6.1.11), it follows that condition (6.1.10)
holds, so the result follows from Theorem 6.1.1.
For equation (6.1.2) we present the following result.

Corollary 6.1.2. Equation (6.1.2) is oscillatory if there exists a set S
1R.o such that
(2.=2)2]
+
limsup t [ps(t) and

lim
t-+oo, tES
iot
i'
t-+oo
q(s)ds
Proof. Because of (6.1.8), !5(M, m)

00.
00
(6.1.30)
(6.1.31)
((7 - 1)/(7 + 1))2. So, the result
Corollary 6.1.3. Assume that

limapproxsup
t -+ 00
fot q(s)ds
io
= 00.
(6.1.32)
Then equation (6.1.2) with "( = 1 is oscillatory.

Proof. If "( = 1 and (6.1.32) holds, then f.k(S) = 00 for some set
S c 1R.o for which (6.1.31) holds. Hence, the result follows from Corollary
6.1.2.
Finally, we present the following result which applies in particular to

functions f(x) of the form f(x) = 2:~=1 cilxl1'i sgn x, where the constants
Ci and exponents "(i are such that
xf(x) > 0 and f'(x);:::: C> 0 for some constant C> 0 and all x
=1= O.
(6.1.33)
Further results on the oscillation of differentia,} equations
463
If we assume 0 < 'Y1 < 'Y2 < ... < 'Yn, then clearly a necessary
condition for (6.1.33) to hold is C1 > 0, Cn > 0 and 'Y1 ~ 1 ~ 'Yn.
Theorem 6.1.2. Assume f(x) satisfies condition (6.1.33) and assume
that condition (6.1.32) holds. Then equation (6.1.1) is oscillatory.
Proof. Since (6.1.32) holds, it follows that the equation xl/(t)+cq(t)x(t) =
is oscillatory by Corollary 6.1.3, where c is the same as in (6.1.33).
Hence, by Theorem 4.3.5 equation (6.1.1) is oscillatory.
6.1.3. Further Results

In the case of equation (6.1.2) the results of Subsection 6.1.1 are fairly
sharp. In fact, first we note that if condition (6.1.31) holds, then this
condition also holds with S replaced by its closure S. We can therefore
state the following partial converse of Corollary 6.1.2.
Theorem 6.1.3. Let S be any closed subset of IRo such that
limsupps(t) <
t....;oo
'Y - 1)2
(- ,
'Y + 1
'Y-=/=1.
(6.1.34)
Then there exists a continuous function q = q(t) with

lim
t....;oo,
tES
10t
q(s)ds =
00
(6.1.35)
and such that equation (6.1.2) has a nontrivial nonoscillatory solution.

Proof. In view of (6.1.34) we can choose EO
t [ps(t) + EO -
(~ ~ ~
r]
Bo,
> 0 and Bo > 0 such that

t E IRo
(6.1.36)
Choose an open set U C IRo such that S C U and p,(U\S) ~ 1 and define
Q1 = IRo\U, Q2 = U\S. Thus, Q1 is closed, Q1nS = 0, IRo = SUQ1UQ2.
Hence, P,(Q2) ~ 1 and so, tpQ2(t) ~ 1 and tpQ, (t) :::: t[l - Ps(t)]- 1.
Define the function pet) by
1+E1, tES
{ 'Y - 1
~' tEQ1
pet) =
where 0 < E1 < 1 will be chosen sufficiently small. It follows that Ip(t)1 ~
2 + Ib -1)/(2'Y)1 on S U Q1, and we can extend pet) to all of IRo so
that it is of class C 1 andsatisfies Ip(t)I~Co=2+lb-l)/(2'Y)1 forall
t E IRo. We now let the function wet) be defined by
wet) = 'Y - 1 pet) - p2(t),

'Y
t E IRo
(6.1.37)
Chapter 6
464
and notice that
t E S
w(t)
(6.1.38)
and
Iw(t)1 :S
1'Y~IICo+c6'
tElRo.
(6.1.39)
For convenience, we set
C1
I"I ~ 11 Co + C6 and A =
We next define u(t) on
E1
(~ + 1 + (1) .
(6.1.40)
lRo by
u(t) = B1
lot w(s)ds,
(6.1.41)
where B1 will be chosen sufficiently large later.

If we write St = S n [0, tJ, QIt = Q1 n [0, t], Q2t = Q2 n [0, t], then
we have
u(t) = B1
r w(s)ds + Jr
1s
w(s)ds +
Qlt
r w(s)ds.
lQ2t
(6.1.42)
It follows from (6.2.38) - (6.2.40) that
1st w(s)ds
= -
r w(s)ds =
JQlt
(~ + A) tps(t),
(6.1.43)
('Y-l)2tPQ1(t)
(6.1.44)
2"1
and
(6.1.45)
Hence, we have from (6.1.42) - (6.1.45) and (6.1.36) after rearranging
(6.1.46)
where
465
and
After some more rearranging, we see that Dl can be written as
Hence, since ,\ can be made arbitrarily small, in view of (6.1.40) it follows

that Dl > 0 and if we choose Bl sufficiently large, we have Do::::: 1.
Therefore, from (6.1.46) we find that u(t) ::::: 1 + Dlt ::::: 1, t E IRo.
Furthermore, since u'(t) = w(t) ::::: C l , t E IR o, we also have u(t) :::::
Bl + CIt, t E IRo. We now define r = r(t) on IRo to be the solution of
the equation
r'(t) = utt)p2(t)r(t),
on IRo \Q2 = S U Ql. We have
p2 (t ) >
C~
u(t) - Bl + CIt'
r(O) = 1
(6.1.47)
{Ill} .
C 2 = min 1, '"'( 2~
(6.1.48)
Since J.t(Q2)::::: 1 it follows from (6.1.47) and (6.1.48) that
lim r( t)
t-+oo
(6.1.49)
00.
We next define v = v(t) by
v(t)
r(t)u(t) > 0,
(6.1.50)
E IRo
then in view of (6.1.49), we have limHoo v(t) = O.

Ixl"Y sgn x, we define the positive function x = x(t) by
= v(t),
f'(x(t))
then, we have
hmt-+oo x(t)
i.e.,
v(t)) l/(-y-l)
= (-
x(t)
'"'(
{Oif'"'(>l
f 0
001
1
<'"'.
With
f(x)
(6.1.51)
Now, by (6.1.41),
(6.1.37), (6.1.9) and (6.1.51), we find
w(t) =
u'(t)
=
('Y ~ 1) p(t) _ p2(t)
f"(x(t))f(x(t)) ()
2( )
(f'(x(t)))2 P t - P t
f"(x(t))f(x(t)) ()
2( )
v 2 (t)
P t - pt.
Chapter 6
466
On the other hand, we also have u(t)
therefore by (6.1.47) and (6.1.51)
,
v'(t)
u (t) = - r(t)v2(t)
r'(t)
r2(t)v(t)
= l/(r(t)v(t
-
by (6.1.50), and
f"(x(tx'(t)
2
r(t)v2(t) - p (t).
x'(t)
f(x(t
(6.1.52)
p(t)r(t).
If we now define q(t) by q(t) = [(p(t) - l)r(t)]', then in view of (6.1.50)

and (6.1.52), we have
q(t)
(p(t)T(t)' _ r'(t) = _ x"(t) + (x'(t2 f'(x(t _ T'(t)

f(x(t
J2(x(t
x"(t)
2
2
,
x"(t)
- f(x(t + p (t)r (t)v(t) - T (t) = - f(x(t'
Thus, x(t) is a positive solution of equation (6.1.2). Furthermore, we have

q(s)ds = [p(t) - l]T(t) - C3 for some constant C3, and hence for t E S
in view of the definition of p(t), we have
J;
(6.1.53)
so that (6.1.49) and (6.1.53) imply that limHoo,tEs
J; q(s)ds =
00.
6.2. Oscillation Criteria for HalfLinear Differential Equations

Here we shall study the oscillatory behavior of the half-linear second
order differential equation
('ljJ(x'(t)'
+ q(t)'ljJ(x(t =
0,
(6.2.1)
where
(i)
q(t) E C(lRo,lR),
(ii) 'ljJ(8) = 181",-18 with a> 0 is a constant.
In Chapter 3, we have shown that some of the oscillatory criteria of

Chapter 2 for second order linear differential equations can be extended
to equation (6.2.1). These criteria are mostly based on the Riccati transformation, since we know that if x(t) is a nonzero solution of (6.2.1),
then
w
() _ 'ljJ(x'(t
t - 'ljJ(x(t
(6.2.2)
Further results on the oscillation of differential equ8,tions
467
solves the generalized Riccati equation
w'(t)
+ q(t) + alw(t)I(<>+l)/<> = o.
(6.2.3)
In these criteria, (6.2.1) is viewed as a perturbation of the nonoscillatory

equation
(6.2.4)
('lj;(x'(t)))' = O.
In the present section, we shall use a different approach which is based on
the relationship between positivity of the 'a + I-degree' functional
I(x;a,b) =
lb
[lx'(s)Ia+ 1
q(s)lx(s)I<>+l] ds
in the class of functions satisfying x(a) = 0, x(b) = 0 and the discojugacy of (6.2.1) in [a, b]. Moreover, we shall investigate (6.2.1) not as
a perturbation of (6.2.4), but as a perturbation of the generalized Euler
equation
(6.2.5)
('lj;(x'(t)))' + t7~1 'lj;(x(t)) = 0,
where 'Yo = (a/(a+l))<>+l is the so called critical constant in this equation.
As in the linear case, if we replace 'Yo by a constant 'Y > 'Yo ('Y < 'Yo),
then equation (6.2.5) becomes oscillatory (remains nonoscillatory).
6.2.1. Preliminary Results

We recall some results from the oscillation theory of linear equations.
The well-known 'variational principle' states that the equation
(a(t)x'(t))'
+ q(t)x(t)
where a, q : [a, b] -* JR, a(t) > 0 is disconjugate in [a, b]'

nontrivial solution has at most one zero in [a, b], if and only if
I(x;a,b) =
lb
(6.2.6)
i.e., any
[a(s)(x'(s))2 - q(s)x2(s)] ds > 0
for every nontrivial, piecewise GI[a, b] function for which x(a) = 0 = x(b).
Another important concept in oscillation theory of linear equations is the
principal solution. A solution Xo of (6.2.6) is said to be principal if
limHoo xo(t)/x(t) = 0 for any nonzero solution x of equation (6.2.6)
which is linearly independent with Xo (solution x is said to be nonprincipa0. A principal solution of equation (6.2.6) exists (uniquely up to
multiplication by a nonzero real constant) if and only if (6.2.6) is nonoscillatory.
Now consider (6.2.6) as a perturbation of the nonoscillatory equation
(a(t)x'(t))'
+ qo(t)x(t)
= 0,
(6.2.7)
Chapter 6
468
where qo is a continuous real-valued function, and let Xo, Xl be principal

and nonprincipal solutions of (6.2.7), respectively. If
lim Xl ((t))
t-H>a
Xo t
1=
t
[q(s) -
qo(s)]x~(s)ds
> 1,
(6.2.8)
then equation (6.2.6) is oscillatory, and if
XI(t)
t~n~ xo(t)
00
+ ?
1
[q(s) - qo(s)] xo(s)ds < 4'
then (6.2.6) is nonoscillatory, here h+(t)
(6.2.9)
= max{O, h(t)}. In particular, if
a(t) == 1, qo(t) = 1/(4t 2 ) then the condition
lim(lnt)l[q(s)-~lsds
t
4s
t--+oo
> 1
(6.2.10)
is sufficient for the equation
x"(t)
+ q(t)x(t)
(6.2.11)
to be oscillatory, and the condition

lim (1n t)
t--+=
1=
t
[q(s) -
~]
4s
sds <
(6.2.12)
is sufficient for the nonoscillation of (6.2.11).

In the following study, we shall show that condition (6.2.10) is extendable to (6.2.1), while there are some difficulties in extending condition
(6.5.12) to half--linear equations. The analysis relies on Lemma 3.2.6 which
we restate here for the convenience of the reader.
Lemma 6.2.1. Let
be a solution of the equation
(a(t)v;(x'(t)))'
on [a, b] satisfying x(t)
+ q(t)~(x(t)) =
0,
(6.2.13)
0 on (a, b). Denote by Y the family
Y = {y(t) E Cl[a,b]: y(a) = 0
y(b)
and
y(t) f 0 on (a,b)}.
Then for every y E Y,
J(y;a,b) =
lb
[a(s)ly'(s)I"+l - q(s)ly(s)I"+l] ds 2 0,
where equality holds if and only if y and
(6.2.14)
are proportional.
A closer examination of the proof of this lemma shows that the condition
[a, b] in definition of Y can be replaced by a wea.ker condition:
y E Cl
469
Further results on the oscillation of differentia'] equations
y is piecewise of the class Cl in [a, b], and at any discontinuous point

[a,b] of y' there exist finite limits y'(t+), y'(t-). This larger class
of function we shall denote as Y1 .
tE
Consequently, if we can find a nontrivial solution y E Y1 such that
I(y; a, b) <:: 0, then (6.2.13) is conjugate in [a, b], i.e., there exists a nontrivial solution with at least two zeros in [a, bj. Conversely, if I(y; a, b) > 0
for every nontrivial y E Y1 , then (6.2.13) is disconjugate in [a,bj. Indeed,
if y is a nontrivial solution with consecutive zeros t 1, t2 E [a, b], then for
we have
to
[a(s)ly'(s)lo+ 1
tl
a(t)y(t)?fJ(y'(t))l
q(s)ly(s)la+lj ds
t2 _ j.t 2y(s) [(a(s)?fJ(y'(s)))' + q(s)?fJ(y(s))] ds =

t1
o.
tl

The main result of this section is the following theorem.
Theorem 6.2.1.
t~~,ont)
Suppose that
where 10 = (0:/(0:
00
[q(s) -
+ 1))<>+1,
S2~1] s"'ds
>
2C:t: 1)
a,
(6.2.15)
Proof. In view of Lemma 6.2.1 it suffices to find for any d > 0 a piecewise
differentiable function x with compact support in (d,oo), say, [to, t 3 ]
such that
Let t3 > t2 > tl > to and let f, 9 be solutions of (6.2.5) satisfying

the boundary conditions f(to) = 0, f(t l ) = t~/(a+1), g(t2) = t~/(<>+l),
g(t3) = O. Define a test function x as follows
x(t) =
~(t{f
i:
<::t:o<:: t <:: tl
ta/(MI) if tl <:: t <:: t2

g(t) if t2 <:: t <:: t3
if
t 2: t 3 ,
470
Chapter 6
and for convenience let
H(t2' t3)
Vo
t~/(Q+I) [Vot;-a/(a+l)
eX: 1)
a,
xo(t)
P(9'(t 2))] ,
-1
ta/(a+I).
Now integration by parts gives
Next, we claim that the functions f /xo and g/xo are strictly monotonic on
(to, tl) and (t2' t3) respectively. If (f /xo)'(f) = 0 for some t E (to, tr),
thenwehave (f'/f)(f) = (xb/xo) (f). Denote by WI = '1j;(f')N(f) , W2=
'ljJ(x~)/'ljJ(xo). Clearly, the functions WI and W2 satisfy the generalized
Riccati equation
W' + ~ + alwl(a+l)/a = 0
(6.2.16)
t a +1
with the same initial condition at t = t. Hence, these solutions coincide

over the whole interval of their existence, which is the interval (to, tl)'
471
But this is an obvious contradiction (since f(to) = 0), i.e., (f /xo)' > 0
on this interval. The same argument applies to prove the monotonicity of
g/xo over (t2J t3).
Note that the second mean value theorem of the integral calculus implies
the existence of a constant (1 E (to, td such that
Similarly,
j ~2
t2
q(s) -
I]
oS
"'~1 s"'ds
for some 6 E (t2' t3). Consequently,
Next, we shall derive the asymptotic formula for G(to, tt) as t1 -+ 00

and for H(t2' t 3) as t3 -+ 00. For this, we use the transformation t = eU.
Put z(u) = x(e U ) = x(t), v(u) = tOw(t), then this transformation transforms (6.2.5) and (6.2.16) into the equations (with constant coefficients)
(4;(z'))' - mj;(z')
and
+ 104;(z) =
(6.2.17)
Vi = av - 10 - alvl(o+l)/", = F(v),
(6.2.18)
respectively. Clearly, equation (6.4.18) possesses the constant solution v =

Vo = (a/(a + 1))0<, and any solution v tends to Vo as u -+ 00. Indeed,
F(vo) = 0 and if v = v(u) exists on some interval [T,oo), then
V (U)
veT)
d(
F(()
Now, if u -+ 00, the integral on the

occurs only if v( 00) = Vo.
(6.2.19)
= v - T.
left~hand
side must diverge and this
The function F(v) can be expressed in the form (since F'(vo) = 0),
F(v) = (1/2)F"(vo)(v - VO)2 + O((v - von as v -+ Vo, and hence
_1_
_ 2 + 0(_1_)
F(v) - F"(VO)(V - VO)2
vo
V -
as
v -+ vo,
Chapter 6
472
this implies
2
as
u- T
F"()(
va va-v ) + O(ln Iv - vol)
u -+
00.
Thus, we have
2
- - + (va - v)O(1n Iv - vol) = (va - v)(u - T)
u -+
as
F"(vo)
00.
Since, limu->oo(vo - v(u))O(ln Iva - v(u)l) = 0, it follows that

lim
u->oo
lim u(vo - v(u)) = ~().

1<->00
F" Vo
v(u))(u - T) =
(1'0 -
Consequently, O(lnlvo-v(u)I)=O(1n(l/u))=O(1nu) as u-+oo, and

thus l/(vo - v) = (1/2)FI/(vo)u + O(ln u) as u -+ 00, which means
v - va
!F"(vo)u + O(1n u)
_
F"(vo)u
~F"(vo)u (1 + 0
(1_0(lnu))
__
F"(vo)u
e:
U ))
+o(lnu)
u2
as t -+ 00. Taking into account the relation between solutions wand v

of (6.2.16) and (6.2.18), we have
o
t w(t) -
1'0
2
(In(lnt))
= - FI/(vo)
In t + 0 ~
as
t -+
00.
Let w(t, to) = 'ljJ(f'(t))j'lj;(f(t)). Then the function w solves (6.2.16), and
hence
as t1 -+ 00, since by direct computation F"(vo) = -l/vo. Moreover,

observe that G(to, t 1) is positive and decreasing. Indeed, if U1 = In hand
v(u) = eOUw(eU, to), then G(to, t 1) = V(U1) - Vo > 0 for every U1 > Uo
(since fWd > x~(td) and v'(u) = F(v(u)) < H(vo) < O.
Concerning the asymptotic behavior of H(t2' t3)
proceed similarly as for G(to, t1), to obtain
o
[va tH(t 2, t)
3 -- to/(a+l)
2
2 /(a+1)
as t3 -+
00,
we
a
t 2a2 /(a+1) W (t 2, t)]
3
-- Vo - t 2 W (t 2, t)
3 ,
473
where w(t, t3) = 'ljJ(g'(tN(g(t is the solution of (6.2.3) generated by the

solution 9 of the equation (6.2.5), i.e., the solution for which w(t:;, t 3 ) =
-00. We claim that Vo - t ZW(t2, t3) -t 0 as t3 -t 00. For this, we use
again the transformation t = eU , v( u) = tOw( t) which trallsforms (6.2.3)
into equation (6.2.18). Let u E IR be arbitrary and denote by w(u, u)
the solution of (6.2.18) determined by the solution of equation (6.2.17)
satisfying z( u) = 0, i.e., lim u -+ u- w( u, u) = -00. Similarly, as for U-t
00 in the previous part of the proof, we have limu-+= ['('0 - t'( u, u)] = O.
Using the fact that (6.2.18) is autonomous, i.e., 'l'(v, - 'u, u) solves this
equation too, we have Vo - v(u, u) -t 0 as lu -ill t 00, regardless
whether u -t -00 and u is fixed, or '11 is fixed and u -t 00.
Consequently, if U2 = Int2, U3 = Int3 and V(U,'113) = t CY W(t,t3), t = ell,
we have 0 = lim U3 --+=[vo - V(U2' U3)] = limt3-+CXl [vo - t W(t2, t 3)], i.e.,
1'0 - t W(t2, t3) -t 0 as t3 -t 00. Now, let d:::; to be fixed, we find
I(x;to,t3)
G(to,tl) +H(t2,t3)
G(to, tl)
:::; G(d,t1) [ G(d,tl)
-1~2
[q(s) -
1~2
H(t2' t3)
+ G(d,t 1 )
G(d,6) ~,
s~~l] s"'ds
1'0 ]
q(s) - S,,+l s ds .
(6.2,20)
Next, we let E > 0 be a sufficiently small number such that the limit
in (6.2.15) is greater than 2(0:/(0: + lCY(l + 60) = 21'0(1 + 6E). We have
limt--+=(lnt)G(d,t) = 21'0, i.e., (Inx)G(d,t) < 2Vo(1+E) if t is sufficiently
large, and in view of (6.2.15), to > d can be chosen in such a way that
G(;6)
1~ [q(s) - S~~l] sCYds
whenever 6 > to and
. G(to,t)
11m
-=,:-:-,--,-
t--+=
G(d,t)
(ln~l) (1~ [q(s) - }~l] S<>dS)
1
)
(In6)G(d,6)
>
21'0 (1 + 6f)
> 1 + 4f,
2vo(1 + f)
is sufficiently small. Since
(6.2.21)
. vo-t"w(t,to)
111n
--'---....:.,-c.....::e-
t-+oo
11m
t--+oo
1'0 - tCYw( t, d)
F"(:::o2)lnt
-2
F"(vo) Int
+ 0 (~)
+0
(In(lnt)
JnTt
there exists a tl > to such that
G(to, t 1 )
< 1 + f.
G(d,tl)
Further, (6.2.21) implies the existence of t2 > tl such that
G(d1 ~ )
,1
1'] s"'ds
6
[q(s) - <>~l
E,
s
> 1 + 3f whenever
1,
474
Chapter 6
Finally, since H(t2' t3) -40 as t3 -4 00, we find that H(t2' t3)C- 1(d, tl) <
if t3 is sufficiently large. Combining the above calculations, it follows
from (6.2.20) that I(x; to, t 3) < -E, and hence I(x; to, t3) < 0, and now
by Lemma 6.2.1 equation (6.2.1) is oscillatory.
Remark 6.2.1.
1. As in Chapter 3, we observe that the transformation of the independent
variable
lot a-1/a(s)ds
(6.2.22)
transforms (6.4.13) into the equation
d~ (~ (:u x) ) + a1/a[t(u)]q(t[u]N1(x)
0,
which is a equation of the form (6.4.1), here t = t(u) is the inverse of

u = u(t) given by (6.2.22). Using this transformation a criterion similar to
that given in Theorem 6.2.1 can also be formulated for the general equation
(6.2.13).
2. In Theorem 6.2.1 and also in the earlier remarks, (6.2.1) is viewed as
a perturbation of the generalized Euler equation (6.2.5). Of course, one
may also consider (6.2.13) as a perturbation of the one-term nonoscillatory
equation (a(t)~(x'(t)' = O.
Now, we shall prove the following oscillation criterion.
Joo a-1/a(s)ds
and
(6.2.23)
00
t~~ (Jt a-1/a(S)ds) (1
00
q(S)dS) > 1,

Proof. For the exact construction of the test function x(t) for which the
functional I given in (6.2.14) is negative, we define
t:::; to
0,
x(t)
t a-1/O(S)dS/ ito(tl a-1/O(s)ds,
ito
tl:::; t :::; t2
1,
t3
a-1/a(s)ds /
1:
a-1/O(s)ds,
to:::; t :::; tl
475
Now, if to < t1 < t2 < t3 are sufficiently large, then as before we have
I(x; to, t3) < O.
Our next result is a 'nonoscillatory supplement' of Theorem 6.2.2.

Theorem 6.2.3. Suppose that condition (6.2.23) holds and
< (0: + 1)<>+1'
(6.2.24)
then equation (6.2.13) is nonoscillatory. Here q+(t) = max{O, q(t)}.

Proof.
We shall show that the hypotheses of the theorem imply the
existence of N E R such that
for any nontrivial C 1 function x with compact support in (N, 00) (see
Lemma 6.2.1). For this, first we shall establish the following inequality:
Let M be a positive differentiable function for which M'(t) =1= 0 in [a, bJ
and let z E Y (Y is defined as in Lemma 6.2.1). Then,
Indeed, using integration by parts and the Holder inequality, we have
hence the required inequality follows. Now, denote
l/
= (0: + 1)<>+1 '
and let N E R be such that the expression in (6.2.24) is less than v for
t > N. Using (6.2.25) and (6.2.24), we find for any differentiable x with
476
Chapter 6
compact support in (N,oo),
LXJ q(s)Jx(s)J"+lds
(a
+ 1)
:::; (a
+ 1)
:::;
oo
q+(S)JX(s)J"+lds
L (J: X'(U)~j;(X(U))dU)
LJx'(s)J~j;(x'(s))M(s) (J8 !;(~~)dU)
L
< (a + l)v
oo
q+(s)
ds
oo
oo
oo
M(s)Jx'(s)J1f;(x(s))ds
<>
:::; (a+1)v(LOOJM'(S)JJx(s)J"+ldS)
< (a + l)"+lv
: :; L=
-
ds
00
<>+1
(LOOJ~~l;~J:IJx'(s)J"+ldS)
1
<>+1
JM(S)J"+l
Jx'(s)J"+lds
JM'(s)J"
a(s)Jx'(s)J"+lds
(since JM(t)J"+l/JM'(t)J" = a-1/"a(t)). Hence, we have
L
oo
[a(s)Jx'(s)J"+l - q(s)Jx(s)J"+l] ds > 0
and this completes the proof.
6.3. Oscillation Criteria via

Liapunov's Second Method
In this section we shall discuss the oscillatory property of solutions of
second order differential equation
(a(t)x'(t))'
+ F(t, x(t), x'(t))
(6.3.1)
by applying Liapunov's second method. In what follows we shall assume

that
(i) a(t) E C([to, 00), lR+), to ~ 0, and

(ii) FE C([to,oo) x lR2 ,lR).
For this, we shall need to consider (6.3.1.) in its equivalent system form
x'(t)
y(t)
a(t) ,
y'(t)
y(t))
-F ( t,x(t), a(t) .
(6.3.2)
477
6.3.1. Preliminaries
Consider the system of first order differential equations
x'(t) = F(t, x(t)),
(6.3.3)
where x is an n-vector and F(t, x) is an n-vector function, F(t, x) E

C([to, oo) x IRn,IR n ), and let Ilxll be the Euclidean norm of x. In what
follows, a Liapunov function will be assumed to be a scalar continuous
function which satisfies locally a Lipschitz condition with respect to x,
i.e., for any compact set KeIRn, there exists a constant L(K) > 0
suchthat IIF(t,xd-F(t,X2)II:::;L(K)llxI-X211 for xl,x2EK.
Let Vet, x) be a continuous scalar function defined on an open set S
and is locally Lipschitzian in x. Corresponding to Vet, x), we define the
function
~~.3.3)(t, x) =
lim sup -hI {Vet

h-.O+
+ h, x + hF(t, x)) - Vet, x)}.
(6.3.4)
Let x(t) be a solution of(6.3.3) which stays in S and denote by V'(t, x(t))
the upper right-hand derivative of Vet, x(t)), i.e.,
V'(t,x(t)) =
limsup -hl {V(t+h,x(t+h))-V(t,x(t))}.

h-.O+
(6.3.5)
For a point (t, x) E S and small h, there exists a neighborhood U of

(t,x) such that 0 c S, (t+h,x+hF(t,x)) E U and (t+h,x(t+h)) E U.
Let L be the Lipschitz constant of Vet, x) with respect to x in 0,
and write
V(t+h, x(t+h)) - Vet, x(t))
= Vet + h, x + hF(t, x) + hE) - Vet, x)

:::; V(t+h, x
where
+ hF(t, x)) + LhliEll -
Vet, x),
(6.3.6)
---+ 0 as h ---+ 0+. From (6.3.6) it follows that
.
1
hmsup -h {Vet
h-.O+
+ h, x(t + h)) -
:::; lim sup .!. {Vet

h-.O+ h
Vet, x(t))}
+ h,x + hF(t,x)) -
(6.3.7)
V(t,x)}.
On the other hand, we have

Vet
+ h, x(t + h)) - Vet, x(t))
~ Vet
+ h, x + hF(t, x)) - LhIlEII- Vet, x),
which implies that

V(6.3.3)(t,X) :::; V'(t,x(t)).
(6.3.8)
Chapter 6
478
Thus, from (6.3.8) and (6.3.7), we obtain
V(6.3.3)(t, x) = V'et, x(t)).
(6.3.9)
By the same calculation, we also find the relation

1
liminf -h {Vet + h, x(t + h)) - Vet, x(t))}

h--+O+
1
liminf -h {Vet + h, x + hF(t, x)) - Vet, x(t))}.
(6.3.10)
h--+O+
If Vet, x) has continuous partial derivatives of the first order, it follows

that
I
V(6.33)(t,X)
av av
at
+ ax . F(t,x),
(6.3.11)
where '.' denotes the scalar product.
Now if V('6.3.3) (t, x) :::::

and consequently V'(t, x(t)) ::::: 0, then the
function Vet, x(t)) is non-increasing in t, which implies that Vet, x)
is non-increasing along a solution of the system (6.3.3). Conversely, if
Vet, x) is non-increasing along a solution of the system (6.3.3), we have
V(~.3.3)(t,x) ::::: 0. Further, if
1
liminf -h {Vet + h, x + hF(t, x)) - Vet, x(t))} ~ 0,
h--+O+
the function Vet, x) is non-decreasing along a solution of the system

(6.3.3), and conversely.
The following property of a Liapunov function V (t, x) is important.
Let xes), yes) be continuous and differentiable functions for s ~ t such
that x(t) = yet) = x. Then by the definition
V'(t,X(t))
.
1
hmsuP -h {V(t+h,x(t+h)) - V(t,x(t))}
h--+O+
and
V'et, yet)) = lim sup -hI {Vet + h, yet + h)) - Vet, y(t))}.
h--+O+
Let L be a Lipschitz constant of V (t, x) in a neighborhood of the point

(t, x). Then for sufficiently small h,
V'et, yet))
:::::
lim sup -hI {Vet + h, x(t + h)) - Vet, yet))}

h--+O+
+ lim sup -hI {Vet + h, yet + h)) - Vet + h, x(t + h))}

h--+O+
< lim sup .!. {Vet + h, x(t + h)) - Vet, x(t))}

h
.
1
+hmsup -hLlly(t + h) - x(t + h)[[.
h--+O+
h--+O+

Thus, we have V'(t, y(t)) :S V'(t, x(t))
479
+ Llly'(t) - x'(t) II.

Theorem 6.3.1. Assume that there exist two continuous functions V(t,
x,y) and W(t,x,y) which are defined on t~T~ta, x>O, Iyl<oo
and t ~ T, x < 0, Iyl < 00, respectively, where T can be large, and
assume that V(t, x, y) and W(t, x, y) satisfy the following conditions:
(i) V(t, x, y) ---* 00 uniformly for x > 0, - 00 < y
W(t, x, y) ---* 00 uniformly for x < 0 and - 00 <
(ii) Vr6.3.2)(t,X(t),y(t)):S 0 for all sufficiently large
is a solution of the system (6.3.2) such that x(t) >
V(~.3.2)(t,
<
00
y <
as t ---* 00, and

as t ---* 00,
00
t, where {x(t),y(t)}
0 for all large t and
x(t), y(t))
.
1
bmsup -h {V(t
h .... a+
+ h, x(t + h), y(t + h)) - V(t, x(t), y(t))} ,
(iii)WC6 .3 . 2 )(t,x(t),y(t)):S0 for all sufficiently large t, where {x(t),y(t)}

is a solution of the system (6.3.2) such that x(t) < 0 for all large t and
W C6 .3.2)(t, x(t), y(t))

=
.
1
bmsup -h {W(t
h .... a+
+ h, x(t + h), y(t + h)) - W(t, x(t), y(t))}.

for t ~ ta. By condition (i), if t is sufficiently large, say, t ~ t1 we find
V(ta,x(ta),y(ta)) < V(t,x(t),y(t))
(6.3.12)
for all x(t) > 0, ly(t)1 < 00 and t ~ t i . However, by condition (ii), we
have V(t,x(t),y(t)) :S V(ta,x(ta),y(ta)) for all t ~ ta, which contradicts (6.3.12). When x(t) < 0 for t ~ ta, by employing the function
W(t,x(t),y(t)) we arrive again at a contradiction. This completes the
proof.
Now consider a Liapunov function v(t, x, y), and define V(6.3.2) (t, x, y)
by
V(6.32)(t, x, y)
li~~~p {v (t + h,x + h art)'Y -
hF (t,x, art))) - V(t,X,y)}.

(6.3.13)
480
Chapter 6
If V(6.3.2)(t,x,y):::; 0, then v(t,x(t),y(t)) is non-increasing in t, where

{x(t),y(t)} is a solution of the system (6.3.2).
To apply Theorem 6.3.1, we shall need the following lemma.
< 0), -00 < y < 00

assume that there exists a Liapunov function v(t, x, y) which satisfies the
following conditions:
Lemma 6.3.1. For all large T ~ to, t ~ T, x> 0 (x
yv(t,x,y) > 0 for y # 0, t ~ T, x > 0 (yv(t,x,y) < 0 for

~ T, :r < 0),
(I2) iJ(632)(t,x,y):::; -k(t), where k(t) E C([T,oo),IR), and
(1 1 )
y
# 0, t
.t
lim inf ;, k(s)ds

t---+oo
T,
for all large
Tl
T.
(6.3.14)
Moreover, assume that there exist a T2 and a w(t, x, y) for all large
such that T2 2: T and w(t, x, y) is a Liapunov function defined
for t ~ T2 , X > 0, y < 0 (x < 0, y > 0) which satisfies the following
conditions:
(13) y:::; w(t,x,y) (-y:::; w(t,x,y)) and w(T2 ,x,y):::; m(y) where m(y)
= 0 and m(y) < 0 (y # 0),
(I4) w(6.32)(t, x, y) :::; -c(t)w(t, x, y) where c(t) E C([T2' (0), IRa) and
is continuous, m(O)
l~ ats) exp (
-l:
C(U)dU) ds =
00.
(6.3.15)
Then if {x( t), y( t)} is a solution of the system (6.3.2) such that x( t) > 0
for all large t, then y(t) 2: 0 (y(t) :::; 0) for all large t.
Proof. Let x(t) > 0 for t 2: to. The case when x(t) < 0 for t ~ to can
be treated similarly. Now assume that there exists a sequence {tn}~=l
such that tn -t 00 as n -t 00. Let tn ~ T and tn be sufficiently large
so that
liminf
t-+CXJ
it
tTl,
k(s)ds
Consider the function v(t,x(t),y(t)) for t
for
~
~ tn.
(6.3.16)
tn. Then, we have
v(t, x(t), y(t)) :::; V(tn' x(t n ), y(tn))
-l~ k(s)ds.
From (6.3.16) it follows that there is To > to such that for all t 2: To,
k(s)ds ~ (l/2)v(t n ,X(tn),y(tn )), because v(tn,x(tn),y(t n )) < o.
Tl~erefore, for t 2: To, we obtain v(t, x(t), y(t)) :::; (1/2)v(tn' x(t n ), y(t n ))
< 0, which implies that y(t) < 0 for all t 2: To.
Itt
481
For To there is a T such that T 2': To and there is a Liapunov

function w(t, x, y) defined on t 2': T, x> 0, y < O. For this w(t, x, V),
we find
y(t) ::; w(t, x(t), y(t))
< w(T, x(T), y(T)) exp (< m(y(T)) exp ( -
J;
ht C(S)dS)
C(S)dS)
for
t 2': T.
Since x'(t) = y(t)/a(t), we have
x'(t) ::; m(y(T)) a~t) exp (-
ht C(S)dS) ,
(6.3.17)
and hence
x(t) ::; x(T)
+ m(y(T))
ht ats)
exp (-
h,s C(U)dU) ds.
Since x(t) > 0 for t 2': T and m(y(T)) < 0, we find by condition (6.3.15)
that x(t) ---+ -00 as t ---+ 00, which is a contradiction. This completes
the proof.
Remark 6.3.1.
replaced by:
When a(t)
= 1 and c(t) = 0, condition (1:l) can be
(13)' ml(y)::; w(t,x,y) and w(T2,x,y) ::; m(y), where ml(y) is

monotone, continuous, mdO) = 0, ml(y) < 0 and m(y) is continuous,
m(O) = 0 and m(y) < 0 (y f 0).
If we can find Liapunov functions which satisfy the conditions of Lemma
6.3.1, then we can prove the following theorem by employing a similar
argument as in Theorem 6.3.1.
Theorem 6.3.2. Let the conditions of Lemma 6.3.1 hold and assume that
for each .5 > 0, there exist a T(.5) > to and Liapunov functions V(t, x, y)
and W(t, x, y) which are defined for all t 2': T(.5), x > .5, y 2': 0 and
t 2': T(.5), x < -.5, y ::; 0, respectively. Moreover, assume that V(t, x, y)
and W(t, x, y) satisfy the following conditions:
(III) V(t, x, y) and W(t, x, y) tend to infinity for all x, y as t ---+
00,
(II 2 ) V(6.32) (t, x, y) ::; 0 as long as V(6.3.2) is defined,
(II3) W(6.3.2)(t,X,y)::; 0 as long a.s W C6 .3.2) is defined.

Since we assume the existence of Liapunov functions satisfying the conditions of Lemma 6.3.1, if x(t) > 0 eventually, then x(t) >.5 eventually
Chapter 6
482
for some IS > 0, because x'(t) = y(t)ja(t)

argument holds for x(t) < 0 eventually.
0 eventually. A similar
As applications of the above results, we consider equation (6.3.1) under

the following assumptions:
(X;; dsja(s) = 00.
(i I )
(6.3.18)
ito
(i2) For t ~ to ~ 0 and x

JI(x) E C(lR,lR) such that
lim inf
t-too
0,
there exist qI(t) E C([to, (0), lR) and
iTt qJ( s )ds ~
for all large
and that xJI(x) > 0, f{(x)

0, lui < 00,
0 for x
qI(t)JI(X)
(i3)
f2(x)
t-too
and that xf2(x)

0, lui < 00,
iTt q2(s)ds ~
> 0,
f~(x) ~
~ to
(6.3.19)
0 and for all large t, x
F(t,x,u).
for all large
0 for x
f=
(6.3.20)
there exist q2(t) E C([to, (0), lR)
For t ~ to and x ~ 0,
E C(lR, lR) such that
liminf
f=
~ to
and
(6.3.21)
0 and for all large t, x :s;
F(t, x, u) :s; q2(t)f2(X).
(6.3.22)
Now, under the above conditions, if {x(t),y(t)} is a solution of (6.3.2)

such that x(t) > 0 for all large t, then y(t) ~ 0 for all large t. For
this, we can assume that (6.3.19) - (6.3.22) hold for all t ~ T* and all
T ~ T*. For t ~ T*, x> 0, Iyl < 00, define v(t,x,y) = ylJI(x). Then,
we have
V(6.3.2)(t,x,y)
n~x)
[-F (t,x, art)) JI(x) -yf'(x) art)]
-qI(t).
Hence, this v(t, x, y) satisfies the hypotheses of Lemma 6.3.1 with k(t) =
qI(t).
Since the condition (6.3.19) implies that for all T ~ T* there is a TI

such that TI ~ T and J;1 qI (s )ds ~ 0 for all t ~ T I , the function
= y + JI(x) J;1 qI(s)ds for t ~ TI , x> 0, y < 0 satisfies the
conditions of Lemma 6.3.1 with c(t) = O. Thus, the conclusion follows.
w(t,x,y)
T*, x < 0, Iyl < 00

andw(t,x,y)=-y-f2(x)J;1 q2 (s)ds, t~Tl, x<O, y>O thenfrom
If we consider functions v(t, x, y) = yj f2(x), t
483
Lemma 6.3.1 it follows that if {x(t), y(t)} is a solution of (6.3.1) such that
x(t) < 0 for all large t, then y(t):::::: 0 for all large t.
Now by employing Theorem 6.3.1 we shall prove the following results.
Corollary 6.3.1. Let conditions (id - (i3) hold. If
00
qi(s)ds =
i = 1,2
00,
(6.3.23)
to

Proof. For t 2: T*, x > 0 and
V(t, x, y)
~ {I
00
h~X) +
t
<
1:
<
set
for
ql(s)ds
for
ql(s)ds
00
y 2: 0
< o.
to
Then clearly V(t, x, y) ---+

and we have
00
00
<y<
00
[-F (t, x, art)) h (x) - yf'(x) art)] + (t)
R~x)
V(~.3.2)(t, x, y)
uniformly for x > 0 and
ql
:::::: -ql (t)
+ ql (t)
0 for t 2: T*, x > 0 and y 2:
o.
Therefore, V(t, x, y) satisfies the conditions of Theorem 6.3.1. Similarly,
W(t, x, y)
{it
q2(s)ds
to
h~x) +
for
y> 0
for
q2(s)ds
y:::::: 0
satisfies the conditions of Theorem 6.3.1. Thus, the conclusion follows.
Corollary 6.3.2. Let conditions (iI) - (is) hold. If
I 1
t
to
00
-1()
qi(u)duds ---+
ass
00
as t ---+
00,
i = 1,2
(6.3.24)
and
00
du
h(u)
<
00
and
du
-00
;
-E
h(u)
<
00
for some
> 0, (6.3.25)
484
Chapter 6
Proof. For
t:::: T*,
V(t, x, y)
x > 0, IYI <
00
du
JI(u)
set
00
rt
+ ito a(s)
00
(6.3.26)
ql(u)duds.
For a solution x(t) which satisfies x( t) > 0 for all large t, we can
assume that x(t) > 0, y(t) :::: 0 for t:::: To :::: to, To is sufficiently large,
and hence
1
JI~~~~ + 1
1
y(t)
- JI(x(t a(t)
V'(t, x(t), y(t
att) [-
00
+ a(t)
00
ql(s)ds
ql(S)dS] .
If we set V*(t,x, y) = -(y/JI(x + f tOO ql(s)ds, then V*(t, x(t), y(t :S

ftOO ql(s)ds, and hence limsuPt->oo V*(t,x(t),y(t :S O. On the other
hand, we have
V'(6_3_2)(t, x, y) = -
R~x) [-F (t, x, art) JI(x) - y!~ (x) art)] -ql(t) :::: O.
Therefore, V*(t, x(t), y(t :S 0,

for
and consequently V'(t, x(t), y(t :S 0
t:::: To.
Similarly, we define W(t, x, y) by
r-
W(t,x,y) = ix
OO
du
h(u)
+ ito
00
a(s)
q2(u)duds.
It is easy to see that W(t, x, y) satisfies the conditions of Theorem 6.3.l.
Thus, the conclusion follows.
We can combine the conditions of Corollaries 6.3.1 and 6.3.2, to obtain

Corollary 6.3.3. Let conditions (it) - (h) hold. If
i 1
t
-1()
to a s s
00
q2(u)duds -+
00
as t -+
00
and
ft"; ql(s)ds =
roo
J-f
du
h(u) <
00,
00,
where E> 0, then equation (6.3.1) is oscillatory.

Corollary 6.3.4. Let conditions ~l) - (i3) hold. If there exist a constant
b> 0 and Pi(t) E Cl([to,oo),lR ), i = 1,2 such that ![(x):::: b for
x
=I 0, i = 1,2
and
485

Proof. In this case for x
y
> 0,
V(t,x,y) = h(X)Pl(t)
+ .fto Pl(S)
and for x < 0, y:<:::; 0
W(t,x,y)
Y
= h(X)P2(t)
+
[ql(S) -
1t [
to
p2(S) q2(S) -
satisfy the hypotheses of Theorem 6.3.1.
1 a( s) (p~ (s) ) 2]
4-bPl(S)
ds
1 a(s) (p~(S)2]
4-bP2(S)
ds
Lemma 6.3.2. In addition to the assumptions of Lemma 6.3.1 assume that

there exists a Liapunov function u( t, x, y) defined for t ~ T*, x > 0, y > r
for r> 0 a large constant, which satisfies
(iid u(t,x,y) -t 00 uniformly for t, x as y -t 00 (as y -t -00) and

u(t,x,y):<:::; f(y) (u(t,x,y):<:::; f(lyl) where f(z) > 0 is continuous,
(ih) u(6.3.2)(t,x,y):<:::; o.
Then if {x(t),y(t)} is a solution of (6.3.2) such that x(t) > 0 (x(t) < 0)
for all large t, then y(t) is bounded for all large t.
Proof. Let x(t) > 0 and y(t) ~ 0 for t ~ T ~ T*. By Lemma
6.3.1 there is such a T. Let K be such that y(T) < K, K > r. There
is a constant c > 0 such that u(t, x, K) :<: :; c, and there also exists an
M > 0 for which we have c < u(t,x,M) for all t ~ T and x> 0 by
condition (iid. But, in view of (ih) we arrive at a contradiction. Hence,
o :<: :; y(t) :<: :; M for all t ~ T.
Theorem 6.3.3. Let the hypotheses of Lemma 6.3.2 hold and assume
that for each 6 > 0 and m > 0 there exist a T(6, m) > 0 and two
Liapunov functions V(t,x,y) and W(t,x,y) such that V(t,x,y) is
defined for t ~ T(6, m), x> 6, 0:<:::; y < m and W(t, x, y) is defined for
t ~ T(6, m), x < -6, - m < y < 0, moreover, assume that V(t, x, y)
and W(t, x, y) satisfy the following conditions:
(III) V(t,x,y) and W(t,x,y) -t
00
uniformly for x, y as t -t
00,
(Ih) V(6.3.2)(t,x,y):<:::; 0 as long as V(~.3.2) is defined,

(Ih) W(6.3.2) (t, x, y) :<: :; 0 as long as W(6.3.2) is defined.
Proof. Let x(t) be an eventually positive solution of (6.3.1). By Lemma
6.3.1 there exists a tl > to ~ 0 such that x(t) > 0, y(t) ~ 0 for t ~ h.
486
Chapter 6
By Lemma 6.3.2 there is a constant m > 0 such that 0:::; y(t) < m for all
t 2: t l . Since x'(t) = y(t)/a(t) 2: 0 for t 2: tI, we have x(t) 2: x(td > 0
for t 2: tI
Consider the Liapunov function V(t, x, y) defined for t 2: T(tS, m), x >
8, 0 :::; y < m where 8 = x(t I )/2, where T 2: t l . The rest of the
proof is similar to that of Theorem 6.3.1. When x(t) < 0 eventually, by
employing the Liapunov function W(t, x, y) and Lemma 6.3.2 we arrive
at a contradiction. Thus, equation (6.3.1) is oscillatory.
As applications of Theorem 6.3.3, we consider the general differential

equation
x"(t)
+ a(t, x(t), x'(tx'(t) + F(t, x(t), x'(t
(6.3.27)
and its equivalent system
x'(t)
y'(t)
=
=
y(t)
- a(t, x(t), y(ty(t) - F(t, x(t), y(t
(6.3.28)
We assume that
(ad F(t, x, y) E C([to, 00) x m?, lR) and xF(t, x, y) > 0 for x i= 0,
(a2) a(t,x,y) E C([to,oo) x lR2 ,lR) and there exist two functions g(t),
p(t) E C([to, 00), lR o) such that - g(t) :::; a(t, x, y) :::; p(t) for t 2: to, x E
lR, y E lR,
(a3) for any 8 > 0 and m > 0 there exist a T(8, m) and a function
q(t; 8, m) 2: 0 defined for t 2: T(8, m) such that limt-+oo J;(cl,m) q(s; 8, m)ds
= 00, and for Ixl 2: 8, Iyl :::; m and xy > 0, IF(t, x, y)1 2: q(t; 8, m),
(a4)
roo
ito
g(s)ds < 00,
lim
t~oo ito
exp (-
itor P(U)dU) ds =
00.
Corollary 6.3.5. If conditions (ad - (a4) hold, then equation (6.3.27) is

oscillatory.
Proof. It suffices to construct Liapunov functions which satisfy the conditions of Theorem 6.3.3.
For t 2: to 2: 0, x > 0,
v(t, x, y)
Iyl < 00
the function
487
satisfies the conditions of Lemma 6.3.1 with k(t) = 0. For any T 2: to,
the function w(t, x, y) = y defined for t 2: T, x > 0, y <
satisfies the
conditions of Lemma 6.3.1. For this, we have
V(6.3.28)(t, x, y) =
a(t, x, y)y - F(t, x, y) :::; - p(t)y :::; - p(t)w(t, x, y)
and I; exp (- I;p(u)du) ds =
= y2 exp ( -2 Itto g(s)ds)

It':' g(s)ds <
,
00
00.
Moreover, it is easily seen that u(t, x, y)
satisfies the conditions of Lemma 6.3.2, since
and F(t,x,y) > O. Furthermore, we can see that
{ (-y)exp (l>(S)ds)
v(t,x,y) -
t 2: to, x < 0, y 2: 0
for
(-y)exp (-
9(S)dS)
for
t 2: to, x < 0, y < 0
w(t,x,y) = -y for t 2: T, x < 0, y > 0 and u(t,x,y) = exp (-2 It: g(s)ds)
for t 2: to, x < 0, y <
satisfy the conditions of Lemma 6.3.2. Next, for
and m > 0 define V(t, x, y) for t 2: T(8, m), x > 8, 0 :::;
each 8 >
y < m by
V(t, x, y) = yexp (where L
t 9(S)dS) + L t
lto
IT(o,m)
q(s; 8, m)ds,
= exp (- It':' g(s)ds) > O. Then, we have
-1:
-1:
V(~.3.28)(t, x, y)
exp (
< exp (
9(S)dS) [-g(t)y - a(t, x, y)y - F(t, x, y)]

9(S)dS) [-g(t)y + g(t)y - q(t; 8, m)]
:::; - (exp (-l: g (S)ds)) q(t;8,m)
+ Lq(t; 8, m)
+ Lq(t; 8, m)
+ Lq(t; 8, m) :::; 0.
Thus, we see that V(t, x, y) satisfies the conditions of Theorem 6.3.3.

Similarly, the Liapunov function
W(t, x, y) = (-y) exp (-it 9(S)dS)

to
+L
rt
q(s; 8, m)ds
IT(o,m)
satisfies the conditions of Theorem 6.3.3. Thus, it follows from Theorem

6.3.3 that equation (6.3.27) is oscillatory.
488
Chapter 6
Corollary 6.3.6.
repla.ced by
In Corollary 6.3.5 conditions (a3) and (a4) can be
(a5) for any fJ> 0 there exist a T(6) > 0 and a q(t;6) > 0 defined for
T(6) such that
t ~
( exp (
and for
Ixl
(a6) lim
t->oo
-1:
~
g( S)ds ) ) t(ln q( s; fJ)ds --+
fJ, xy::::: 0, IF(t, x, y)1
00
t --+
as
(6.3.29)
00
q(t; 6), and
ltot exp (- ltot P(U)dU) ds = 00.
(6.3.30)
Proof.
From conditions (al), (a2) and (a5) it follows that there are
Liapunov functions which satisfy the conditions of Lemma 6.3.1 as was
seen in Theorem 6.3.2. For t ~ T(6), x> 6, y ~ 0 define
V(t,x,y) = yexp
(-it q(S)dS) +
to
Then, we have
V(~.3.28)(t,x,y)
exp
(-1:
< exp (
T(6), x < -6, y
W(t, x, y)
q(s;6)ds.
+ q(t; fJ)
9(S)dS) [-g(t)y + g(t)y - q(t; 6)]
+ (ex p
~
IT(8)
9(S)dS) [-g(t) t(8) q(s; fJ)ds
-g(t) (exp
For t
to
9(8)dS) [-g(t)y - a(t,x,y)y - F(t,x,y)]
-1:
-1:
(-1:
-1:
+ exp (
(expit -9(S)dS)
9(S)dS)) t(8) q(s;fJ)ds
9(S)dS) ) q(t; fJ)
O.
0 if we define W(t, x, y) by
(-y) exp ( -l>(S)ds) + (ex p ( - l : g (S)dS)) t(8/(S; 6)ds,
then we also have W(6.3.28)(t,x,y) ~ O.

Therefore, by Theorem 6.3.2
From Corollaries 6.3.5 and 6.3.6 the following corollary is immediate.

Corollary 6.3.7.
replaced by
In Corollary 6.3.5 conditions (a3) and (a4) can be
489
(a7) given 8> 0 there exists a T(8) > to and a q(t; 8) ;::: 0 for t;::: T(8)
with
-1
t
lt
T(Ii)
q(s; 8)ds -+
as
00
t -+
00
and such that for Ixl;::: <5, xy > 0, IF(t, x, y)1 ;::: q(t; <5),
(as) for some To;::: to and a constant lvI,
~ exp (l>(S)dS)
< M
for
t;::: To
and condition (6.3.30) holds.

1. The results of Section 6.1 are taken from Butler and Erbe [3]. Corollary 5.1.3 is a known result of Olech et. a1. [8]. Theorem 6.1.1 can easily
be extended to more general equations of the form
(a(t)x'(t))'
and
(a(t)x'(t))'
+ q(t)f(x(t)) =
+ q(t)lx(t)I'"Y
sgn x(t) = 0,
where a(t) E C([to,oo),lR+). For more details see Butler and Erbe [3].
2. The results of Section 6.2 are taken from Dosly [5]. The method used
in Theorems 6.2.1 and 6.2.2 as well to establish (6.2.8) and (6.2.9) suggests
the following general approach to the investigation of oscillatory properties
of equation (6.2.13). In this method, this equation would be viewed as a
perturbation of the general nonoscillatory equation
(a(t)1f;(x'(t)))' + qo(t)1f;(x(t)) = O.
(6.4.1)
However, except for the special case treated in Theorem 6.2.1, no 'halflinear' analogy of (6.2.8) and (6.2.9) is known. The main reason is the
absence of equivalents of principal and non principal solutions, and also one
does not have a transformation theory similar to that for linear equation.
Using the linear transformation theory, for example, criterion (6.2.8) may
be proved as follows:
If Xo is the principal solution of equation (6.2.7), then the transformation x = xou transforms this equation into the one term equation
(a(t)x5(t)u'(t))'
0,
(6.4.2)
hence this transformation converts (6.2.6) into the equation
(a(t)x6(t)u'(t))'
+ [q(t) -
qo(t)]x6(t)u(t)
O.
490
Chapter 6
Now, the last equation is treated as a perturbation of (6.4.2) in a way as

suggested in Remark 6.2.1(2) (applied to the linear case) and the obtained
results are then transformed 'back' into equation (6.2.6). This approach
cannot be directly extended to half-linear equations because of the absence
of a 'half-linear' transformation theory. In fact, this is the main difficulty
in extending condition (6.2.12) to half-linear equation (6.2.1).
Therefore, it would be interesting to obtain results which can fill the gap
between the theory known for linear equations and that for the half-linear
equations.
3. The results of Section 6.3 are taken from Yoshizawa [9,10J. Similar
results for the difference equations are available in Agarwal and Wong [IJ.
The obtained results offer alternative proofs for some known results. In
fact, Corollaries 6.3.1 - 6.3.7 include results due to Bobisud [2], Coles [4],
Macki and Wong [6J and Opial [7J. It is interesting to proceed further in
this direction and establish via Liapunov second method oscillation criteria
for differential and functional differential equations of arbitrary order.
6.5. References
1. R.P. Agarwal and P.J.Y. Wong, Advanced Topics in Difference Equa-
tions, Kluwer, Dordrecht, 1997.

2. L.E. Bohisud, Oscillation of solutions of damped nonlinear equations,
SIAM J. Appl. Math. 19(1970), 601-606.
3. G.J. Butler and L.H. Erhe, A generalization of Olech-Opial-Wazewski
oscillation criteria to second order nonlinear equations, Nonlinear Analysis
11(1987), 207-219.
4. W.J. Coles, Oscillation criteria for nonlinear second order equations, Ann.
Mat. Pura. Appl. 83(1969), 123-134.
5. O. Dosly, Oscillation criteria for half-linear second order differential equations, Hiroshima Math. J. 28(1998), 507-521.
6. J.W. Macki and J.S.W. Wong, Oscillation of solutions to second order
nonlinear differential equations, PaciHc J. Math. 24(1968),111-117.
7. Z. Opial, Sur les integrales oscillantes de l'equation diffrentielle u" +
f(t)u = 0, Ann. Polon. Math. 4(1958), 308-313.
8. C. Olech, Z. Opial and T. Wazewski, Sur Ie probleme d'oscillation des
integrales de l'equation y" + g(t)y = 0, Bull. Akad. Polon. Sci. 5(1957),
621-626.
9. T. Yoshizawa, Stability Theory by Liapunov's Second Method, The Mathematical Society of Japan, Tokyo, 1966.
10. T. Yoshizawa, Oscillatory property for second order differential equations,
Tohoku Math. J. 22(1970), 619-634.
Chapter 7
Oscillation Results for
Differential Systems
7.0. Introduction
In this chapter we are concerned with the oscillation of nonlinear twodimensional differential systems and second order vector-matrix differential
equations. In Section 7.1 we shall present criteria for the oscillation of nonlinear two-dimensional differential systems. This includes the superlinear,
linear, and sublinear cases. Section 7.2 deals with the oscillation of linear
second order differential systems. Here, first the system considered will be
reduced to a certain scalar Riccati inequality, so that the known results
from the literature can be applied to obtain oscillation criteria. Then, we
shall employ the notation and definitions of Section 6.1 to present some general results. Finally, we shall use Riccati and variational techniques which
involve assumptions on the behavior of the eigenvalues of the coefficient matrix (or of its integral) to present a number of sufficient conditions which
guarantee the oscillation of linear second order systems. In Section 7.3 we
shall discuss the oscillation of nonlinear second order differential systems
with functionally commutative matrix coefficients. Here, we shall show
that the oscillation theory of such systems can be effectively reduced to
the study of diagonal systems of scalar second order differential equations.
In Section 7.4 we shall prove some comparison theorems of Hille-Wintner
type for second order operator-valued linear differential equations. In Section 7.5 some oscillation results for second order differential systems with
a forcing term are given.
7.1. Oscillation of Nonlinear TwoDimensional Differential Systems

The oscillation of second order nonlinear differential equations is of particular interest and, therefore, it is widely studied in the literature. A
Chapter 7
492
challenging problem is to extend known oscillation criteria for second order

nonlinear differential equations to nonlinear two-dimensional differential
systems. Such differential systems include in particular second order nonlinear differential equations.
Consider the nonlinear two-dimensional differential system
x'(t)
ql(t)h(y(t))
-q2(t)!2(X(t)),
v'et)
(7.1.1)
where
(i) qi(t) E C([to,oo),JR), i = 1,2 and ql(t) 2:: 0 for t 2:: to,
(ii) fi(X)EC 1 (JR,JR) and satisfy xfi(x) >0 for x~O, i=1,2, hex)
is increasing on JR and f~(x) 2:: 0 for every x ~ O.
In what follows, we shall restrict our attention only to the solutions of
the differential system (7.1.1) which exist on some ray [To, 00), where
To 2:: to may depend on the particular solutions. Of course, under quite
general conditions existence of such solutions can always be guaranteed. As
usual, a continuous real-valued function defined on an interval [To,oo) is
said to be oscillatory if it has arbitrarily large zeros, and otherwise, it is
said to be nonoscillatory. A solution (x, y) of the system (7.1.1) is called
oscillatory if both x and yare oscillatory, and otherwise, it is called
nonoscillatory. The differential system (7.1.1) is said to be oscillatory if an
its solutions are oscillatory. We remark that if the coefficient ql(t) is not
identically zero on any interval of the form [TO, 00) where TO 2:: to, then
from the first equation of (7.1.1) for any solution (x, y) the oscillation of
x implies the same for y. So, if (x, y) is a nonoscillatory solution of
(7.1.1), then x is always nonoscillatory.
The special case where fi(X) = Ixl>'i sgn x, x E JR (Ai> 0), i = 1,2 is
of particular interest. In this case the differential system (7.1.1) becomes
x'(t)
ql(t)ly(tW" sgn yet)

-Q2(t)lx(tW'2 sgn x(t),
v'et)
where Ai, i
of (7.1.1).
= 1,2
(7.1.2)
are positive constants. System (7.1.2) is the prototype
In the particular case where Ql(t) > 0 for t 2:: to and hex)
JR, (7.1.1) reduces to the equation
Cl~t) Xl(t)), + Q2(t)f2(X(t))

Further, when Ql(t)
== 1 for
xl/(t)
O.
x, x
(7.1.3)
2:: to, equation (7.1.3) becomes
+ Q2(t)f2(X(t))
O.
(7.1.4)
493
Oscillation results for differential systems
The prototype of equation (7.1.4) is the so called Emden-Fowler equation
xl/(t)
+ q2(t)lx(tW
= 0,
sgn x(t)
(7.1.5)
where A2 > O. System (7.1.2) reduces to (7.1.5) when q1(t) == 1 for

t ::::: to and A1 = 1.
7.1.1. Superlinear Case

It has been shown (see Theorem 4.1.3) that for equation (7.1.5) if A2 >
1, then a sufficient condition for oscillation is
(7.1.6)
Moreover, it is well-known that if q2(t) ::::: 0 for t::::: to, then condition
(7.1.6) is also necessary for t.he oscillation of equation (7.1.5). Our first
t.heorem generalizes this result to the system (7.1.1). For t.his, we define
R(t) = ftto q1(s)ds and assume that
lim R(t) =
t -+co
(7.1. 7)
00,
and 12 satisfies the strong superlinearity condition, namely,
CO
du
h(u) <
00
and
du
-CO
12(u) <
00.
(7.1.8)
TheoreIll 7.1.1. Suppose conditions (7.1.7) and (7.1.8) hold. If q2(t) ::::: 0
for t::::: to and
00
R(S)q2(S)ds
00,
(7.1.9)
to
then the system (7.1.1) is oscillatory.

Proof.
Assume that the system (7.1.1) admits a nonoscillatory solution
(x, y) on an interval [to, (0), to ::::: o. From condition (7.1.7) it follows

that q1 (t) . 0 on any interval of the form [To, (0), To ::::: to. So, as
remarked above, x(t) is always nonoscillatory. Without loss of generality,
we assume that x(t) '" 0 for all t::::: to. Furthermore, we observe that the
substitution z
= -x, v = -y transforms (7.1.1) into the system
A(v( t))
Z' (t)
q1 (t)
v'(t)
-Q2(t)i2(z(t)),
where ii(u) = - ii( -u), u E lR, i = 1,2. The functions ii, i = 1,2 are
subject to the conditions posed on ii, i = 1,2. Thus, we can restrict our
494
Chapter 7
discussion only to the case where x is positive on [to,

either one of the following two cases holds:
(I}) y(t)::; 0 for t 2:: to,
or
00 ),
and assume
(12) y(t) 2:: 0 for t 2:: to.
(h) Let y(t) ::; 0 for t 2:: to. The second equation of (7.1.1) implies
that y(t) is decreasing on [to, (0) and y(t) -+ either - 00, or a finite
negative number as t -+ 00. Similarly, it follows that h (y(t)) -+ either
- 00,
or finite negative number as t -+ 00. This in view of condition
(7.1.7) implies that
q}(s)h(y(s))ds = -00. Now, integration of the
first equation of (7.1.1) gives
It':
x(t)
= x(to) +
t qds)h(y(s))ds
-+ -
1to
as t -+
00
00,
which contradicts the assumption that x(t) > 0 for t 2:: to.
(1 2 ) Let y(t) 2:: 0 for t 2:: to. Define w(t) = h(y(t))/h(x(t)), t 2:: to.
Then as a consequence of the system (7.1.1) the following equation holds
w'(t)
+ q}(t)f~(X(t))W2(t) + q2(t)ff(y(t)) =
(7.1.10)
O.
Clearly, q}(t)f~(x(t))w2(x(t)) 2:: 0 for t 2:: to, and since x(t) > 0 and
y(t) 2:: 0 for t 2:: to, y(t) decreases to a nonnegative constant. Also there
is a positive constant k such that
f{(u) 2:: k > 0
for
We thus have
w' (t) ::; - kq2 (t)
[O,y(to)].
(7.1.11)
for t 2:: to
(7.1.12)
Multiplying both sides of (7.1.12) by R(t) and integrating from to to t,

we get
R(t)w(t) Since,
to
t q}(s)w(s)ds
1to
q}(s)w(s)ds =
it
to
::; - k
x'(s)
f ( ( ))ds
2 X S
t R(S)q2(S)ds.
(7.1.13)
lto
=
(t)
x(to)
du
-(f) <
2 U
kIt:
and R(t)w(t) > 0 for t 2:: to, we find that

R(S)q2(S)ds <
contradicts condition (7.1.9). This completes the proof.
00,
00,
which
Next, we shall extend Theorem 4.1.1 (without condition (4.1.11) (see

alternative proof after Lemma 4.1.3) to (7.1.1). For this we shall assume
that
{CO Q2(s)ds exists as a finite number,
lto
(7.1.14)

and define Q(t)
technical lemma.
= ft q2(s)ds,
OO
495
2: to. We shall also need the following
Lemma 7.1.1. Let (x, y) be a solution on [to,oo), to 2: 0 of the system

(7.1.1) with x(t) > 0 for all t 2: to. Moreover, let T 2: to and c be a
real constant. If
y(t o)
- h(x(to))
ft
fT,
f~(x(s))
+ ito Q2(s)ds + ito y(s)x (s) fi(x(s)) ds 2:
for every t 2: T, then y(t)::::: -ch(x(T)) for all t 2: T.

Proof. From the second equation of (7.1.1) we find for t 2: T,
ito
y'(s)
ds
f2(x(s))
y(t)
y(to)
- h(x(t)) + h(x(to)) -
t
' f~(x(s))
ito y(s)x (s) fi(x(s)) ds
and so, we have
y(t)
f2(x(t))
y(to)
[ - f(x(to))
'
ft
f~(x(s))
fT,
+ ito Q2(s)ds + ito y(s)x (s) fi(x(s)) ds

f~(x(s))
+ iT y(s)x (s) fi(x(s)) ds
for every
2: T.
This for all t 2: T, by the hypothesis gives
t [- h(x(s))
y(s)] [ , f~(X(S))]
-x (s) h(x(s)) ds.
y(t)
h(x(t)) 2: c + iT
Now by Lemma 3.1.1, we conclude that w(t)::::: -y(t) for every t 2: T,

where w satisfies
w(t)
h(x(t))
ft
w(s)
f~(X(S))]
+ iT h(x(s)) -x (s) h(x(s)) ds
for t 2: T.
Since w'(t) = 0 on [T,oo), we have w(T) = ch(x(T)). Thus, w(t) =

ch(x(T)) for all t 2: T. This completes the proof.
Theorem 7.1.2. Let the function JI

00
/
du
JI 0 12 (u) <
h(u)
-00
00
and
JI(h(u)) satisfy
du
JI 0 h(u)
<00
(7.1.15)
and suppose that conditions (7.1.7) and (7.1.14) hold, and for all sufficiently
small u and every v> 0,
JI(u)JI(v) ::::: JI(uv) ::::: JI(u)(-JI(-v)).
(7.1.16)
496
Chapter 7
If
00 ql (s)h (Q(s))ds
(7.1.17)
00,
to

Proof. As in the proof of Theorem 7.1.1 without loss of generality, we
assume that x(t) > 0 for t 2': to. From the first equation of (7.1.1) it
follows that the function y(t)x/(t) is necessarily nonnegative for a.ll t 2': to,
even though y(t) is oscillatory. First, we claim that
00
to
f'(x(s))
y(s)x/(s) f~(x(s)) ds <
(7.1.18)
00.
For this, assume that (7.1.18) is not satisfied. By condition (7.1.14) there
exists a constant k such that
y(to)
- Jz(x(to))
lt
to
>
q2(s)ds - k for t 2': to
Furthermore, we can choose a point tl 2': to such that
tl
to
f'(x(s))
y(s)x/(s) f~(x(s))ds > 1- k.
So, for every t 2': tl, we have
y(to)
- h(x(to))
lt
to
q2(s)ds+
itl /
to
fHx(s))
y(s)x (s) n(x(s))ds 2': 1
and hence, by Lemma 7.1.1 with T = tl and c = 1, we obtain y(t) ~ b

for all t 2': tI, where b = - Jz(x(h)) ~ O. Next, from the first equation
of (7.1.1) we have for t 2': tl,
lt
x(t) - x(td =
ql(s)h(y(s))ds
t,
h(b)
lt
ql(s)ds,
t,
which in view of condition (7.1.7) gives limt-400 x(t) =

tion. Thus (7.1.18) holds.
-00,
a contradic-
Now in view of (7.1.14), the definition of Q and (7.1.18), the second

equation of (7.1.1) for t 2': to, gives
Q(to) - Q(t)
lt
to
q2(s)ds = -
lt
to
y/(s)
Jz(x(s)) ds
y(t)
y(to)
- Jz(x(t)) + h(x(to)) -
00
it /
to
f' (x(s))
y(s)x/(s) fhx(s)) ds,
f~(x(s))
y(s)x (s) n(x(s))ds
Oscillation results for differentia'] systems

i.e.,
y(t)
h(x(t)) = 8 + Q(t)
497
rOO
f~(x(s))
+ it y(s)x (s) J?(x(s)) ds
(7.1.19)
for every t 2': to, where the real number 8 is defined by
100
y(to)
8 = f2(x(to)) - Q(to) -
to
f~(x(s))
y(s)x (s) J?(x(s)) ds.
We claim that the constant 8 is nonnegative. Otherwise, from (7.1.14)

and (7.1.18) it follows that there exists T 2': to such that
rOO
iT
f~(x(s))
y(s)x (s) J?(x(s)) ds :s; -
roo
it
and
q2(s)ds <
for all t 2': T.
Thus, by using (7.1.19), we find for every t 2': T,
y(to)
- h(x(to))
lt
to
q2(s)ds +
IT
I
f~(x(s))
y(s)x (s) fi(x(s)) ds
to
00
f~(x(s))
- 8 - Q(to) - . to y(s)x (s) J?(x(s)) ds

- 8-
100
t
Q2(s)ds -
100 y(s)x (s) n(x(s)) ds 2': -8 + -8 + -8 = --8

I
f2(x(S))
and so Lemma 4.3.1 ensures that y(t):s; b1 for all t 2': T, where b1 =
(8/2)h(x(T)) < O. Hence, exactly as in establishing (7.1.18) we arrive at
the contradiction limt-.oo x(t) = -00, which proves the claim.
Finally, (7.1.19) guarantees that y(t) 2': Q(t)f2(x(t)) for every t 2': to.
Hence, in view of limHoo Q(t) = 0 and the condition (7.1.16), from the
first equation of (7.1.1), we obtain for t 2': to,
and consequently
du
(t)
x(to)
h(u)
2':
it
to
Ql(s)h(Q(s))ds
for all
2': to
So, because of (7.1.15), we have
to
Ql(s)h(Q(s))ds :s;
100
x(to)
f ; ( ) <00
1
2 U
for
2': to,
498
Chapter 7
Example 7.1.1. Consider the differential system

x'(t)
yl/3(t)
y'(t)
sin t
- ( --+
t
for t 2': to = 1f/2. Here, we have ql(t)

hey) = yl/3 and hex) = x 5 . Now,
Q(t)
roo
= it
q2(s)ds
roo
= it
(7.1.20)
2 + cos t) 5()
x t
t2
= 1, q2(t) = (sint/t)+(2+cost)/t2 ,
-d
(2 +
2+cost
coss)
s
and hence Q(to) = Q(1f/2) = 4/1f exists as a real number. Also, we find
that F(u) = h 0 h(u) = u 5 / 3 which is a superlinear function and satisfies
Joo u- 5 / 3du < 00. Finally,
= i t (2+COSS)1/3 dS -+
tq1(s)h(Q(s))ds
ito
7r/2
00
as
t-+oo.
Thus all conditions of Theorem 7.1.2 are satisfied, and hence the system
7.1.2. Sublinear Case

For equation (7.1.5) it is known that if q2(t) > 0 for t 2': to and
0< >'2 < 1, then a necessary and sufficient condition for oscillation is that
Jt"; sA2q2(S)ds = 00. In this subsection, this result will be generalized to
the system (7.1.1).
Theorem 7.1.3. Suppose q2(t) 2': 0 for t 2': to and condition (7.1.7)
holds. In addition, assume that fI(x) 2': 0 for x#- 0, i = 1,2 and for all
v 2': 0 and u > 0,
-h(-u)h(v) 2': h(uv) 2': h(u)h(v)
and the composite function
namely,
1
du
io hoh(u)
If
<
00
00
(7.1.21)
satisfies the sublinearity condition,
<
and
q2(s)h(R(sds =
00.
00,
to
where R(t) = Jt: ql(s)ds, then the system (7.1.1) is oscillatory.
(7.1.22)
(7.1.23)
499
Proof. First we shall show that for any fixed tl > to,
(7.1.24)
Since limHoo R(t) = 00, there is a t2
> tl such that
R(t) ;::: max{l, 2R(td}
t > t2.
for
(7.1.25)
Thus, we have
(7.1.26)
and hence by condition (7.1.21) for t > t2,
h(R(t) - R(td) ;::: h(R(t)/2) ;::: h(1/2)h(R(t)).
(7.1.27)
Now (7.1.24) follows from (7.1.27) and (7.1.23).

Suppose now that the system (7.1.1) is not oscillatory, so that x(t) and
y(t) are of fixed sign for t > t l . Using (7.1.24), we may use a translation
to shift tl to to. Thus we may assume without loss of generality that
x(t) > 0 and y(t) is of fixed sign for all t;::: to.
As in the proof of Theorem 7.1.1 the two possibilities (h) and (12) are
considered, and the case (II) can be disposed of in exactly the same way as
in Theorem 7.1.1. Thus, we consider
(b) Assume that y(t);::: 0 for t;::: to. From the system (7.1.1) we note
that x(t) is an increasing function on [to, 00) while y(t) is a decreasing
function on [to, 00). Integrating the first equation of (7.1.1) from tl to
t, we obtain
x(t) = x(to)
lto
ql(s)h(y(s))ds;:::
lto
R(t)fI(y(t)).
ql(s)h(y(s))ds
(7.1.28)
Thus, for all sufficiently large t such that R(t);::: 1,
h(x(t)) ;::: h(R(t))fI(Yl(t)) ;::: h(R(t))h
fI(y(t)).
Using the second equation of the system (7.1.1), we get

xh(R(t))h 0 fI(y(t)), or
y'(t)
fI(y(t)) > q2(t)h(R(t))
for
t;::: to
(7.1.29)
y'(t) S; -q2(t)
(7.1.30)
Y (t O)
y(t)
du
2
f ( );:::
1 U
it
to
q2(s)h(R(s))ds.
(7.1.31)
Chapter 7
500
The right-hand side of (7.1.31) tends to 00 as t --+ 00, while the left-hand
side of (7.1.31) remains bounded by (7.1.22) as y(t) --+ 0 as t --+ 00. This
contradiction completes the proof.
Example 7.1.2.
Consider the system
x'(t)
y3/5(t)
y'(t)
r14/9x5/9(t),
2 to
1.
Here we have ql(t) = 1, fJ2(t) = C 14 / 9, h(y) = y3/5 and 12(1:)

Clearly, R(t)=J:ds=t-1 and !2oh(U)=U 1 / 3 . Now
(7.1.32)
= x 5/ 9
Thus all conditions of Theorem 7.1.3 are satisfied, and hence the system
7.1.3. Further Oscillation Criteria

The main result here is the following.
satisfies
lim inf Ih (u) I > O.
(7.1.33)
'U---*(X)
If
lim
t .....
=lto
fJi(s)ds =
00,
= 1,2
(7.1.34)

Proof. Without loss of generality we assume that x(t) > 0 for t 2
to. Dividing the second equation of the system (7.1.1) by h(x(t)) and
integrating over [to, t], we obtain
_ltot h(x(s))
y'(s) ds
t q2(s)ds
lto
= Q*(t).
(7.1.35)
Integrating by parts the first term of (7.1.35), we get
y(t)
- h(x(t))
Q (t)
+ -
y(to))
h(x(to))
+ lto
,f~(x(s))
y(s)x (s) fi(x(s)) ds. (7.1.36)
The last term of the equality (7.1.36) is nonnegative as f~(x) 2 0 and

y(s)x'(s) = fJl(s)y(s)h(y) 2 0, s 2 to. Further, in view of (7.1.34) there

exists a tl 2': to such that Q*(t)
Thus, we have
501
+ (-y(to)/h(x(to))) 2': 1
t ( -x (s) /1(x(s))
fHX(S)))
(-y(s))ds
y(t)
- h(x(t)) 2': 1 + it1
for all t 2': h.
for all
2': t 1
(7.1.37)
Hence, in particular y(t) <
for t 2': tl. Now, Lemmas 4.1.1 and 4.4.1
allow us to compare -y(t) with the solution u(t) of the integral equation
u(t)
h(x(t))
1+
to conclude that
f~(X(S)))
it1rt (-x,(8) n(x(s))
(u(s))ds
-y(t) 2': u(t)
for all
for all
2': h (7.1.38)
t 2': t 1 .
(7.1.39)
Differentiating (7.1.38), we find that u'(t) = 0, t 2': tl and hence u(t) =

1, t 2: t 1 . Thus (7.1.39) reduces to y(t) :::; -1 for all t 2: t 1 . From the
hypothesis (7.1.33), we have JI(y(t)):::; SUPy:S;-l JI(y) = k < 0, where k
is a constant. Hence, by (7.1.34) it follows that
It
Ql(8)JI(y(s))ds :::;
t,
kIt
ql(s)ds -> -
00
as
t ->
00.
t,
Finally, integration of the first equation of (7.1.1) gives
x(t) = X(tl) -
t ql(s)JI(y(s))ds.
it l
(7.1.40)
The fact that the right-hand side of (7.1.40) tends to - 00 as t -> 00

contradicts the assumption x(t) > 0 for t 2': to. This completes the
proof.
Remark 7.1.1. In all the results proved above some positivity condition
has to be imposed on one of the coefficients qi(t). To see this, we will
construct an example of a nonoscillatory system (7.1.1) for which condition
(7.1.34) is satisfied.
We consider the linear case, namely, when Ji(u) = u for i = 1,2.

The new variable w(t) = -y(t) satisfies the Riccati differential equation
w'(t) = q2(t) + ql(t)W 2(t), t 2: to = 0, or upon integration, the Riccati
integral equation
(7.1.41)
-
where Q(t) = w(to) + fo q2(s)ds. The question whether the system (7.1.1)
has a solution x(t) of fixed sign (say positive) is equivalent to the question
whether (7.1.41) has a continuous solution on [0, (0).
Chapter 7
502
In what follows we shall construct step functions Q(t) and q2(t) so that
(7.1.41) has a solution on [0,00). We shall allow qi(t) to change sign, even
though condition (7.1.34) is satisfied. Although our smoothness condition
on qi(t) precludes such step functions, and the solution has jumps at each
of the points 3k, 3k + 1, modification of our example by smoothing out
the abrupt jumps easily leads to an acceptable counterexample.
_{-1,tE[3k,3k+1)
.
rt
_
1, t E [3k + 1, 3k + 3). Then, hmHoo Jo ql(s)ds - 00.
Let qi(t) -
Let us now construct Q(t). In each [3k, 3k + 1), we choose Q(t) to be a

constant a = a (k) so large that a + J03k ql (s )w 2(s) ds = k. Later, we are
going to show that the number J;k ql(S)w 2(s)ds < 0 for all k. Hence,
a(k) 2 k -+ 00
as
k-+oo.
(7.1.42)
In the interval [3k,3k + 1), (7.1.41) reduces to w(t) = k - J:k w 2 (s)ds,

which has the solution w(t) = [t-3k+(1/k)]-1. Note that the right-hand
side of this function is well-defined as the denominator does not vanish in
[3k, 3k + 1). From (7.1.41), we have
k
- - - a < O.
k+1
(7.1.43)
In [3k+1,3k+3), we take Q(t) to be negative of the number in (7.1.43).

Hence, we find that limHoo Q(t) = 00. In [3k + 1,3k + 3), equation
(7.1.41) reduces to the simple equation w(t) = J:k+l w2(s)ds, which has
the trivial solution w(t) = 0 for t E [3k + 1, 3k + 3). Thus,
(7.1.44)
We assumed that the above inequality is true when the upper limit of
integration is 3k and concluded that the same is true for 3(k + 1).
This induction step allows us to infer that (7.1.44) holds for all k > O.
Continuing this process over each interval, we obtain a solution w(t)
defined on the whole half-line [0,00), and so we have an example of a
nonoscillatory system (7.1.1) for which condition (5.3.34) is satisfied.
It is natural to ask whether some relaxation on the positivity requirement is possible. As shown by the above example, it is not sufficient to
assume that at each t one of the q/s is nonnegative.
Remark 7.1.2. We note that Theorem 7.1.4 holds without the assumption
that II is increasing on JR. Clearly, the increasing character of II on
JR implies condition (7.1.33).

Example 7.1.3.
503
Consider the differential system
- tanh y(t)
x'(t)
y' (t)
- [-v'tsint+
2~(2+cost)]
(7.l.45)
Ix(t)I'\ sgn x(t)
t 2: to = Jr/2, A> 0 is a constant. Here, we have ql(t) = l/t,

q2(t) = -v'tsint + (1/2v't)(2 + sint), t 2: to, !l(u) = tanhu, 12(u) =
IvY sgn u, U E JR. For every t 2: to,
for
j q2(s)ds =
t
to
} 1f /2
[-VS sin s+ 1;;; (2 + sin S)] ds =

2yoS
v't(2 + cost) - 2VJr/2 ---+
t ---+
as
00
} 1f /2
d[VS(2+cos s)]
00.
All conditions of Theorem 7.1.4 are satisfied, and hence the system (7.1.45)
is oscillatory. However, Theorems 7.1.1 - 7.1.3 are not applicable.
7.2. Oscillation Theorems for

Linear Differential Systems
Consider the linear second order differential system
x"(t)
+ Q(t)x(t) =
(7.2.1 )
0,
where x: [to, (0) ---+ JRn and Q(t) is a continuous real symmetric n x n
matrix function for t 2: to 2: O. In 1980, Hinton and Lewis conjectured
that (7.2.1) is oscillatory whenever
lim Al
t~oo
(jt Q(S)dS)
to
(7.2.2)
00,
where Al (.) denotes the maximum eigenvalue of the matrix. In partial

answer to this conjecture the following theorem is obtained.
Theorem 7.2.1. If condition (7.2.2) holds and either
liminf
~t
tr
t~oo
or
liminf
t~oo
where tr()
oscillatory.
tr
(jt Q(S)dS)
to
> -
(jt j8 Q(U)dUdS)
to
to
(7.2.3)
00,
> _
00,
(7.2.4)
represents the trace of the matrix, then system (7.2.1) is
504
Chapter 7
We recall some pertinent definitions and notation which will be subsequently used. For any n x n matrix A, the transpose will be denoted
by A*, similarly x* denotes the transpose of the column vector x. If
h, t2 E [to, 00), tl =1= t2 and if there exists a nontrivial solution of (7.2.1)
which vanishes at hand t2, then tl and t2 are said to be (mutually)
conjugate relative to (7.2.1). System (7.2.1) is said to be disconjugate on an
interval J c [to,oo) if every nontrivial solution of (7.2.1) vanishes at most
once in J and system (7.2.1) is said to be oscillatory if for each tl > to
there exists a t2 > tl such that (7.2.1) is not disconjugate on [tl, t2].
The matrix differential system associated to (7.2.1) is
X"(t)
+ Q(t)X(t)
0,
[to,oo)
(7.2.5)
where X(t) is an n x n matrix and Q(t) is as in (7.2.1). A solution

of the system (7.2.5) is said to be nontrivial if detX(t) =1= 0 for at least
one t E [to, 00) and a nontrivial solution X(t) is said to be prepared or
self-conjugate if
X*(t)X'(t) - (X*(t))'X(t)
0,
[to, 00)
(7.2.6)
(we note that for any solution X(t) of system (7.2.5),
X*(t)X'(t) - (X*(t))'X(t)
C,
(7.2.7)
where C is a constant matrix). System (7.2.5) is said to be oscillatory if

the determinant of every nontrivial prepared solution vanishes on [T, 00)
for each T > to. This is equivalent to oscillation of (7.2.1) since any
solution of (7.2.1) is of the form x(t) = X(t)c for some constant vector c
and some nontrivial prepared solution X(t) of the system (7.2.5).
If A is a real symmetric n x n matrix, then its eigenvalues Ak(A),
k = 1,2, ... ,n (which are all real) will be assumed to be ordered so that
(7.2.8)
and
(7.2.9)
k=l
We denote by S the linear space of all n x n real symmetric matrices. A

linear functional cP: S -t 1R is said to be positive if cP(A) 2: 0 for A E S
and A 2: 0, (i.e., A is symmetric positive semidefinite).
In the following discussion, we shall show how condition (7.2.2) leads to
certain scalar Riccati inequality. We will then employ some recent results
on the oscillation of scalar equations to establish a set of conditions weaker
than (7.2.3) and (7.2.4) under which (7.2.2) leads to the oscillation of the
system (7.2.1).
Oscilla.tion results for differentia.l systems
505
For a given solution X(t) of (7.2.5) we define a matrix-valued function

W(t) = -X'(t)X-1(t). This function satisfies the Riccati equation
W(t)
It:
W(to)
where Ql(t) =
Q(s)ds.
obtain the scalar equation
+ Ql(t) +
W2(s)ds,
(7.2.10)
Taking traces on both sides of (7.2.10), we
= tr W(to) + tr Ql(t) +
tr W(t)
ltD
ltD
tr W 2(s)ds.
(7.2.11)
Now suppose condition (7.2.2) holds, i.e., limt-;oo Al(Ql(t)) = 00. Then,
Al(W(tO)+Ql(t))--+oo as t--+oo, and given 1'>0 we can find a large
T 2:: to such that
(7.2.12)
Hence, for all t 2:: T,
tr W 2(t) =
LA%(W(t)) 2:: Ar(W(t)) 2:: (1- E)2Ar (Ql(t)). (7.2.13)

k=l
We use this estimate in (7.2.11) to derive an inequality for the scalar quantity tr W(t). For this, we split the integral term in (7.2.11) as follows
ltD
tr W2(s)ds =
I'
{t tr W2(s)ds + (1 - E)
ltD
ltD
tr W2(s)ds.
(7.2.14)
Assuming t 2:: T and ignoring the contribution from the interval [to, T] to
the last integral, we estimate the remainder by means of (7.2.13), to obtain
(7.2.15)
In the first integral we use the simple estimate tr W 2(t) 2:: 2:~=1 w~k(t) 2::
(l/n)(tr W(t))2, where Wkk(t) is the kth diagonal entry of W, to get
the inequality
(7.2.16)
Thus, combining equations (7.2.11), (7.2.14) with inequalities (7.2.15) and
(7.2.16), we find
tr W(t) 2:: tr Ql(t)+tr W(to)+(1-E)3
t Ar(Ql(S))ds+~n ltDt (tr W(s))2ds.
lT
(7.2.17)
Cbapter 7
506
This inequality holds for all t 2': to. Using condition (7.2.2) we get for all
sufficiently large t 2': TI 2': T,
Hence, if we let 8 = (1 - E)4,

tr W(t), t 2': to and
q(t)
tr QI(t)
replace
81
by
>'i(QI(S))ds,
and define w(t)
nE
t 2': to
(7.2.18)
to
then w( t) satisfies the scalar Riccati inequality
w(t) 2': q(t)
Elt w
2(s)ds
for
t 2': T I .
(7.2.19)
to
Next, we state some known sufficient conditions which ensure that the
inequality (7.2.19) is incompatible with the existence of a continuous function w(t) on the entire half-line [to, 00), to 2': O. For t 2': to, we let
Q1(t) = ft~ Q(s)ds and Q2(t) = ftto QI(s)ds.
(id There exists a measurable subset J of [to,

such that
lim q(t) = 00.
00)
with mes J =
(7.2.20)
tE J, t-->oo
(i2) There exists a mea.surable subset J of [to,

such that (7.2.20) holds, a.nd
q'Y(s)ds
00
00)
with mes J <
00
(7.2.21 )
00
for some constant I E (0,1).

(h) There exists a weight function p E F1 such that
. . f
I1m
1Il
ftto
Hoo
p(s)q(s)ds
t
fto
p(s)ds
> -
00,
>
t _ to
(7.2.22)
where F1 is the family of weight functions which includes polynomial

functions that are strictly positive on [to, 00), and
lim
t-->oo
It
to
[(q(s) - >.)+fds
for every>. E JR, where (4)(t))+
= max{4>(t),O}.
00
(7.2.23)
507
(i4) There exists a weight function p E Fo C F1 such that (7.2.22) holds

and
p(s)q(s)ds
hm sup -=O,t,----00.
(7.2.24)
. I:
Hoc
Ito p( s )ds
Now we shall apply these criteria to present oscillation results for (7.2.1).
Theorem 7.2.2. Suppose condition (7.2.2) holds. Then system (7.2.1)

is oscillatory if for some 8 E (0,1) the function q(t) defined by (7.2.18)
satisfies one of the following conditions:
(II) The set J = {t E [to, (0) : q(t) 2': O} has infinite measure.
(h) The set J = {t E [to, (0) : q(t) 2': O} has finite measure and
q'Y(s)ds = 00 for some 'Y E (0,1).
(13) Conditions (7.2.22) and (7.2.23) hold for some p E Fl.
(14) Conditions (7.2.22) and (7.2.24) hold for some p E Fo.
IJ
Proof. Suppose condition (7.2.2) holds. Then equation (7.2.1) is oscillatory or there exists a function w(t) which is defined and continuous
on the half-line [to, (0) such that inequality (7.2.19) holds for all sufficiently large t. The later possibility is excluded by each of the criteria (Ii)'
i = 1,2,3,4.
The following corollaries are immediate.
Corollary 7.2.1. Let the function q(t) in Theorem 7.2.2 be replaced by

= tr Q1 (t) + ct for t 2': to, where c is any positive constant. Then
the conclusion of Theorem 7.2.1 holds.
q* (t)
Corollary 7.2.2. Suppose condition (7.2.2) holds. Then equation (7.2.1)

is oscillatory if for some constant c, tr Qdt) 2': -ct on a set of t with
infinite measure.
is oscillatory if
liminf
t-+oo
~ tr Q2(t) > -
00.
(7.2.25)
Proof. If condition (7.2.25) holds, then there exists a constant c

such that (l/t)tr Q2(t) + (1/4)ct 2': 0 for t 2': to. Hence,
lit
>0
1
(trQ1(s)+cs)ds 2': -ct.
The result follows from Corollary 7.2.1 (take

(7.2.24.
p(t) = t in (7.2.22) and
508
Cha.pter 7

is oscillatory if
for some integer m 2: 1.

Proof. The proof is same as that of Corollary 7.2.3 (take p(t)
tm).
In our next result we shall use the notation and definitions of Section
6.1, and will need the following lemma.
Lemma 7.2.1. Let V(t) be a continuous n x n symmetric matrix and

limapproxsup _(1) [tr (V(t)
t --+ 00
9 t
1t
to
V 2(S)dS)]
L <
00,
(7.2.26)
where g(t) is positive, absolutely continuous, and non decreasing function

on (to, 00). Then,
limapproxinf -1()
t--+oo
gt
1t
to
>'l(V2(s))ds <
(7.2.27)
00.
Since V(t) = V*(t) it follows that V 2(t) 2: 0, and so F(t) =

ftto V 2(s)ds satisfies F(t2) 2: F(h), to :s: h < t2 < 00, (i.e., F(t2) F(tl) is nonnegative definite). Therefore, since [>'l(V(t)W :s: >'1 (V2(t)) :s:
tr(V2(t)) :s: nAl(V2(t)), it follows that if
Proof.
limapproxinf -1()
t --+ 00
9 t
then
1
limapproxinf -()
t --+ 00
9 t
1t
to
We define H(t) = tr
1t
tr(V2(s))ds
(Itt
to
Al(V2(s))ds
00,
lim approx inf _(1) tr

t --+ 00
9 t
(1 V2 (s )dS)
t
to
o V2(s)ds)
00.
- Mg(t), t 2: to where M > L is
arbitrary. Then, limapproxinfHooH(t)/g(t) = 00 so that the inequality

H(t) > g(t) holds on a set E with f.1(E) = 00, where f.1(E) denotes
the Lebesgue measure of E. Also, if we define the set El by
El =
{t:
g~t) [tr V(t) + H(t)] 2: o} ,
509
then by (7.2.26) we find that P(E1) < 00; since M > L and so the
complement of E1 relative to [to, 00) which we denote by Ef, satisfies
p(ED = 00, and H(t) < -tr V(t) on E'( Since p(E n ED = 00 we
see that the inequality g(t) < H(t) < -tr V(t) holds on En E'( Thus,
:s:
g2(t)
:s:
H2(t)
Ef.
holds (a.e.) on E n
.!.
1t
+ Mg'(t))
< H'(t) +Mg'(t) on EnE1c.
H2(t) -
H2(t)
n tr (V2(T)) = n(H'(t)
and so, we have
< H'(t) +M g'(t)
n -
:s:
(tr V(t))2
H2(t)
g2(t)
(7.2.28)
But, this is a contradiction since
1
t
to
H'(S)
H 2( S ) ds
+M
to
g'(S)
~() ds
9
<
(XJ u- 2 du + M
roo
u- 2 du
19(t o )
1 H(to)
- + g(to)
-- <
H(to)
00
whereas the integral on the left-hand side is infinite since p(E n ED =

00.
Now, we shall prove the following result:

Theorem 7.2.3. Let g(t) be positive, absolutely continuous, and nondecreasing on [to, 00) and
limapproxinf _(1)tr
t---+oo
gt
(1 Q(S)dS) -l >
t
to
-00,
l is a constant, (7.2.29)
00
and that
lim .A1
t-+oo
(1 Q(S)dS)
t
00.
to
(7.2.30)
(7.2.31)

Proof. If system (7.2.5) is not oscillatory, then there exists a nontrivial
prepared solution X(t) of system (7.2.5) with det X(t) -=1= 0 on [h,oo)
for some t1 ?': to. Let V(t) = X'(t)X-1(t), t ?': to, so that V(t) =
V* (t). Clearly, V (t) satisfies the matrix Riccati equation
_V'(t) = Q(t)
+ V2(t),
t?,: h.
(7.2.32)
Integrating equation (7.2.32) from t1 to t, we obtain
-V(t)
+ V(t1) =
1t
t1
Q(s)ds +
1t
t1
V2(s)ds
(7.2.33)
Chapter 7
510
and hence
-V(t)
+ V(t 1 ) 2: it Q(s)ds.
(7.2.34)
t,
In view of the subadditivity of Al and (7.2.34), we have Al (Jtt, Q( s )ds)
Al (- V(t))
+ Al (V(tl))'
Since Al (Jtt, V2(S )ds) 2: Al (- V(t))2,
that limt-tooAl(-V(t)) = limt-tooAl(V2(t))

(7.2.29) and (7.2.33) it follows that
lim
~p~r~ sup
[gtt) (tr (V(t)
lim~p~r~sup
00.
we find
Now, by conditions
[t V (S)dS) ) ]
2
[gtt) (tr (V(td-[tQ(S)dS))] =k-l <
00,
where k = limt-too(1/g(t)) tr(V(td). Therefore, by Lemma 7.2.1, we get

1 it Al(V2(s))ds
9 t t,
limapproxinf - ()
t --+
00
Again from (7.2.34), we have Al
(J/, Q(s)d.s)
<
00.
~ 2>'l(-V(t)) for all large
t (since Al(-V(t)) -+ 00 as t -+ 00) and therefore,

(7.2.35)
for all large t. Thus, by integration, we obtain
(7.2.36)
for all large t, and now taking lim approx inf on both sides of (7.2.36)
as t -+ 00 we get a contradiction to condition (7.2.30). This completes
the proof.
Remark 7.2.1.
implies that
If g(t)
in Theorem 7.2.2, then condition (7.2.2)
so that Theorem 7.2.3 includes Theorem 7.2.1 with

lim a p p rox inf.
The following example is illustrative.
liminf
replaced by
511
Example7.2.1. Let 8>17>0 and 0-=17+0-)/2)-1, k=(17+8)/2.

Define Q(t) by
-ktO"]
[ 0
-ktO"
where q(t)
-00
= 8t b - I -17t'1-I, t 2:: to =
as t --+
g(t) =
Then, gCt)
-q(t)
00
and
)11
[AI
too
Q(U)dU)
00
dS) -y
as t --+
where
= t'1 -
t b --+
Let
00.
0 < 'Y < 1.
as t--+oo and g(t) rvt(2'1+Ih. Wehave
1 it tr (Q(s))ds
g(t)
1. Then, I: tr(Q(s))ds
(lIt Q( s )ds) = t'1 --+
(it (is
.
(
t'1 - t b > - to--Y
2'1+1 )
,
t(2'1+ Ih -
r ~ [l (~, (l
which is bounded below as t --+
9!') l [~, ([ Q{U)dU)
rv
00
provided 8(217 + 1) 2:: T Now
d,
Q{U)dU)) '
d'l
->
00
as t --+ 00. By Theorem 7.2.3 oscillation is guaranteed if 0 < 17 < 8 < 217+ 1
(choose 'Y = (8/217) + 1). If 8> 1, we have (l/t) I: tr(Q(s))ds --+ -00,
so we cannot deduce oscillation by Theorem 7.2.1.
In our next result we will relax condition (7.2.2), and replace hypotheses
(7.2.29) and (7.2.30) of Theorem 7.2.3 by a condition involving the relative
rates of the growth of the largest and smallest eigenvalues of It: Q(s)ds
as t --+ 00. For this, we denote and order the eigenvalues of It: Q(s)ds
as AI(t) 2:: A2(t) 2:: ... 2:: An(t).
We shall need the following lemma.
Lemma 7.2.2. Let p(t) be locally bounded, nonnegative and measurable
on [to, 00) and nonzero a.e. Let q(t) be nonnegative and locally integrable
such that p(t) 2:: q(t) It: p2(s)ds for almost all t 2:: to. Then for all
sufficiently large T 2:: to, q E .c 2 [T, 00).
Proof. Let P(t) = It: p2(s)ds. Then, P(t) is absolutely continuous
and by hypothesis, P(t) > 0 for t 2:: tI, say. We have P'tt) = p2(t) 2::
q2(t)p 2(t), and hence PI(t)/ p 2(t) 2:: q2(t). Integrating this inequality
from T>tl to t, weobtain (1/P(T))-(1/P(t))=I;q2(s)ds, andso
I; q2(s)ds::; l/P(T), proving the lemma.
Chapter 7
512
Theorem 7.2.4. Let one of the following set of hypotheses hold:
(I) (i)
limt~r~inf >'l(t) =
(II) (i)
lim~pJ::r:sup
A1(t)
00,
lim~pJ::r:sup I ~:~~~ I > 0;
and (ii)
limt~r~inf I~:~~~ I> o.
and (ii)
= 00,
Then system (7.2.1) is oscillatory.

Proof.
Assume that (I) or (II) holds and suppose that (7.2.1) has a
nonoscillatory solution X(t). Let W(t) = -X'(t)X-1(t) so t.hat W(t)
satisfies
(t) = W(t)
-It
(t Q(s)ds + C,
W2(s)ds =
t 2> t1 2> to
ltD
t,
(7.2.37)
It is known that for any continuous, symmetric matrix-valued function

a continuously varying orthonormal system may be selected [44J. Thus, we
may choose a locally integrable vector y(t) with Ily(t)11 = 1 such that
(1:
y*(t)
Q(S)dS) y(t)
(7.2.38)
Adt).
Let the eigenvalues of W(t) be J.l1(t) 2> J.l2(t) 2> ... > J.ln(t). By
the preceding remark, we may select a system of (orthonormal) locally
integrable eigenvectors ei (t), i = 1, 2, ... ,n such that
W(t)ei(t) = J.li(t)ei(t),
1 if i
0 if i
ei (t)ej(t) = lS ij =
= 1,, n, to
s, t <
Define the functions Ci(S, t), i
<::::
00
I'= j.j
(7.2.39)
by
y(t)
L Ci(S, t)ei(s).
(7.2.40)
i=l
Clearly, Ci(S,t), i = 1,,n are the projections of y(t) on to the

orthonormal system {ei(s)}i=l' and are locally integrable with respect to
both sand t. We have
n
W(s)y(t)
L J.li (s )Ci(S, t)ei (s).
(7.2.41)
i=l
From (7.2.37) - (7.2.41), we find

n
y*(t)(t)y(t) = t;J.li(t)c7(t,t) -
t (
t;J.l7(s)c7(s,t))
ds.
(7.2.42)
513
At this point, in order to give a clearer presentation of the arguments,

we shall concentrate on the case n = 2. Introduce angle functions , ()
by defining
cos(s)el(tO) + sin(s)e2(to),
- sin (s )e1 (to) + cos (s )e2 (to);
Cl(tO, t) =
cos ()(t),
C2(tO, t) = sinB(t).
(7.2.43)
(7.2.44)
From (7.2.43) and (7.2.44), we find that Cl(S, t) = cos[B(t) - (s)], C2(S, t)
= sin[()(t) - (s)]. If we put o:(t) = ()(t) - (s), we may write (7.2.42) as
-1:
y*(t)if>(t)y(t) = P,l(t) cos 2 o:(t) + P,2(t) sin2 o:(t)

-l>i(S) cos 2[o:(t)+(t)-(s)]ds
p,~(s) sin2[o:(t)+(t)-(s)]ds.
(7.2.45)
Now y*(t)<P(t)y(t) = Al(t) + y*(t)Cy(t). Since C is a constant matrix
and Ily(t)11 = 1, we have
limapproxinf
t ---+ 00
or
y*(t)if>(t)y(t)
limapproxsup
t ---+ 00
00,
(7.2.46)
00
(7.2.47)
y*{t)if>(t)y{t)
according as hypothesis (I{i, or (II(i holds.

We shall demonstrate the incompatibility of (7.2.45), and (7.2.46) or
(7.2.47). Part (ii) of hypotheses (I) and (II) imply that there exists a /j
with 0 < /j < 1 such that
Al(t) I > /j for all t
I A2(t)
where 51
51,
(7.2.48)
c [to, 00) satisfies

p,(51) = 00
if
(I)
holds,
(7.2.49)
if
(II)
holds.
(7.2.50)
and 5'1 = [to, 00) - 51 satisfies
p,{5'1) <
00
Let A(t) = max{IA1(t)l, IA2(t)I}. For t E 51 we have A(t)
:s: A1(t)//j.
514
Chapter 7
For any x
ffi.n with Ilxl! S; 6/4, we find
[y(t) + x]*iI>(t)[y(t) + x]
=
y*(t)iI>(t)y(t) + x*iI>(t)x + y*(t)iI>(t)x + x*iI>(t)y(t)
>
)'1 (t)
>
Adt)-~A1(t)-1166Adt)-(1+~6rIICII
- 2Allxll - A(t) IIxl1 2 - (1 +
~O)
IICII
(7.2.51 )
if tE51
> 3A1(t) - 211CII

Recall y( t) = cos B( t )e1 (to) + sin B( t)e2 (to). It is easy to see that we
may choose x = i(t) with Ili(t)11 S; 0/4 such that y(t) + i(t) =
cosB(t)e1(tO) +sinB(t)e2(to), where B(t) = B(t) + (6/2). Now define x(t)
by
x(t)
={
0 if la(t)1 :::: 6/8 (mod 7f) and la(t) - (7f/2)1 :::: 0/8 (mod 7f),
x(t) otherwise,
and let y(t) = y(t) + x(t). If 5(t) is defined by y(t) = cos B(t)edto)
sinB(t)e2(t), 5(t) = B(t) - (t), then we have
A
la(t)1 ::::
8'
(mod 7f),
15(t) -
~I
::
(mod 7f)
(7.2.52)
for all t:::: to. Now (7.2.45) and (7.2.51) give
y*(t)iI>(t)j)(t)
J-Ldt) cos 2 5(t)
-it J-L~(S)
-it J-L~(S)
+ J-L2(t) sin 2 5(t)
cos 2 [5(t) + (t) - (s)] ds
to
sin 2 [a(t) + (t) - (s)]ds
(7.2.53)
to
::::
1
3A1(t) - 211ctl,
t E 51'
Part (i) of the hypotheses of the theorem yields the existence of a subset
52 of [to, (0) such that
1
3A1(t) - 211CII > 0 for

and
t E 52
(7.2.54)
J-L(5:jJ <
00
if (I) holds,
(7.2.55)
J-L(52) =
00
if (II) holds,
(7.2.56)
515
where 8 2 = [to, 00) - 8 2. Now (7.2.50), (7.2.52) - (7.2.56) show that there
exists a subset 8 = 8 1 n 8 2 of [to, 00) with M(8) = 00 such that
M1(t) cos2 &(t)
+ M2(t) sin2 &(t) - (t Mi(s) cos2[&(t) + >(t) -
ltD
- ltDt M~(S) sin [&(t) + >(t) - >(s)]ds

2
where
O! (
> 0 for
t E 8,
t) is bounded away from 0 and 1f / 2 (mod 1f). Let
UI(t) = M1(t) cos 2 &(t) -
ltD
Mi(s) cos 2 [&(t)
+ >(t)
>(s)]ds
- >(s)]ds
(7.2.57)
(7.2.58)
and let J 1 = {t: U1(t) > O}. By (7.2.52) there exist 01 > 0 and a positive
integer m such that
1
Iwl:S; -
(mod 1f).
(7.2.59)
For i = 1, 2, ... , m define
Ai = {t 2: to : i m- 1 :s; >( t) < mi (mod 1f) }

If t E Ai, we have from (7.2.59) that
U1(t) :s; fi,i(t) - 01
ltD
and fi,i (t)
= { IM1 (t) I, t
Ai
0, otherwIse.
(fi,i(S))2ds.
(7.2.60)
By Lemma 7.2.2 either Mi(t) = 0 a.e. or if qi(t) is defined by fi,i(t) =

qi(t) (fi,i(S) )2ds, we have qi E 2 [T, 00) for T(2: to) sufficiently large.
In the latter case the set {t 2: to : qi (t) 2: od has a finite measure,
and in either case, the set {t E Ai : U1 (tn has finite measure. Since
[to, 00) = U~lAi' we find M(81 ) < 00. Similarly, if
It:
U2(t)
M2(t) sin 2 &(t) -
(t M~(S) sin2[&(t)
ltD
+ >(t) -
>(s)]ds
and h = {t: U2(t) > O}, then M(82 ) < 00. But this contradicts (7.2.56),
and this completes the proof of the theorem for the case n = 2.
For the general case n > 2, the basic idea of the proof is the same.
We introduce the orthogonal matrix U(s) whose rows are the eigenvectors
ei(s) of W(s) (see (7.2.39)). Denote the vector (C1(S, t),, cn(s, t))*
by c(s, t). Assuming without loss of generality that U(to) = I, where I
is the identity matrix, then (7.2.40) gives
c(s, t)
= U- 1(s)c(to, t).
(7.2.61)
Chapter 7
516
Now denote U-I(s) by V(s), the components of V(t)c(to,t) by viet)

and the components of [V(s) - V(t)]c(to, t) by 'Wi(S, t), i = 1,2" ", n.
From (7.2.42) we may write y*(t)cp(t)y(t) as
1:
tJ-Li(t)V;(t) =
tJ-L;(S)[Vi(t)
+ 'Wi(S, t)fds.
(7.2.62)
Replacing yet) by an appropriate yet) and obtaining the corresponding

functions viet), Wi(S, t), we find
y*(t)cp(t)y(t)
tJ-Li(t)V;(t)-1: tJ-L;(S)[Vi(t)+Wi(S' tWds > 0, t E S
(7.2.63)
where IVi (t) I 2 60 > 0 on Sand S is a subset of [to, 00) of infinite
measure, Wi(S, t) is the ith component of [V(s) - V(t)]c(t o, t) and
Ilc(to, t)11 = 1. Note that (7.2.63) is the analog of (7.2.53) and (7.2.56).
Since the n x n orthogonal matrices (identified with zn-l) form a
compact set, we may find a finite decomposition of zn-l, {Edk=l' say,
such that G, H E Ek imply IIG - HII < 60 /2, k = 1,2"", m. Now
define Ak = {t 2 to: Vet) E Ed. Then we shall have
[Vj(t) + Wj(s, tW 2
whenever s, tEAk, k
~65,
(7.2.64)
= 1,2, ... ,m, j = 1,2, ... ,n.
With (7.2.63), (7.2.64) and Lemma 7.2.2, we may complete the proof as
in the case n = 2.
Example 7.2.2. We define the 3 x 3 matrix Q(t) by specifying its

integral from 0 to t as follows
lot Q(s)ds
=
tl/2 cos2 t - t n sin 2 t(1
+ cos t) ~tl/2 sin2t[1 + tn-leI + cost)]
0]
~tl/2sin2t[1 + t n -
+ cost)]
(1
t l / 2 sin2 t - t n cos 2 t(I
+ cos t)
where n 2 1. A straightforward computation shows that the eigenvalues

of J~ Q(s)ds are A,l(t) = t l / 2, A,2(t) = 0, A,3(t) = -tn[I + cost]. Let
tm = (2m + I)1l' and t = tm + s. Then it is easily verified that for t near
to t m , 1A,3(t)1 ::; CIS 2 m n for some constant CI. Since 1A,I(t)1 2 c2m l / 2
for some constant C2, for t near t ro , we have
, A,l(t) , > 1 if
A,3(t) -
lsi < c m(1/4)-(1l'/2)

-
C3
517
Thus, {t: IA1(t)I/IA3(t)1 ~ I} has measure at least C4 2:::=1 m(1/4)-(7f/2)

for some constant C4 > 0, i.e., has infinite measure if n::::; J /2. It follows
that hypothesis (I) of Theorem 7.2.4 holds, and hence equation (7.2.1) is
oscillatory. It is easily verified that Theorem 7.2.3 is not applicable for
2::::; n ::::; 5/2.
Now we shall present some sufficient conditions so that a.ll prepa.red
solutions X(t) of (7.2.5) are oscillatory. Once again the results obtained
are based on Riccati techniques. We shall need the following lemma.
Lemma 7.2.3. Assume equation (7.2.5) is nonoscillatory on
Then a necessary and sufficient condition that
lim
T-+oo
IT
t
W2(s)ds
[to, 00).
(7.2.65)
exists for any solution W(t) = -X'(t)X-1(t) of (7.2.10), where X(t) is

a prepared solution of (7.2.5) is that
L(Q)
litiS
liminf -
t-+oo
to
tr Q(u)duds >
to
(7.2.66)
00.
Proof. Suppose first that the limit in (7.2.65) exists for some solution
W(t) = -X'(t)X-1(t) of (7.2.10) where det X(t) =/= 0 for t ~ to. Then
from (7.2.10), we have
tr W(t)
where C
+ tr
00
[tr W(s)
: :; ~ [1:
-1
00
Ito;' W 2(s)ds.
tr W 2(U) dU
[(tr W(S2
(1
and since (tr W(t2 ::::; n tr W2(t),

00. Similarly,
t -+
1: (1
00
tr Q(s)ds - C,
(7.2.67)
to
= -tr W(to) - tr
1:
1t
W2(s)ds =
00
Since
ds
tr W 2(U)dU)
we find (l/t)
tr W2(u)du
2] dS]
(7.2.68)
It: tr W2(s)ds -+ 0
ds -+ 0 as t -+
as
00
so that (7.2.67) and (7.2.68) imply
~
t
it [1
to
to
tr Q(u)du -
c]
ds -+ 0
as
t -+
00.
(7.2.69)
Chapter 7
518
Now since in view of Cauchy-Schwarz inequality
it follows that
lim ~
t-+oo t
Jtot [Jtot
tr Q(U)dU] ds
exists
(7.2.70)
so that (7.2.66) holds.

Conversely, suppose (7.2.66) holds and let Wet) = -X'(t)X-1(t),
where X(t) is a prepared solution of (7.2.5) with det X(t) 01 0, t 2 to.
lt
1
--
to
W(s)ds+-1
t
lt 1
5
to to
l I l t1
5
W2(s)duds--(t-to)W(to)-t
to to
Q(u)duds
(7.2.71)
and hence in view of (7.2.66) it follows that
11t
-t
to
tr W(s)ds
11t 1
+-
to
/'0
tr W2(u)duds ~ M
for some M > O. Since tr W 2(t) 2 0 for t 2 to, limt-+oo
Itto tr
exists and is finite or infinite. Suppose that

t --+ 00. Then,
11 l
t
to to
tr W2 (u )duds --+
00
as
It:
From (7.2.72) it follows that (1ft)

tr W(s)ds --+
for large t again from (7.2.72), we find
-1
t
Itto l
to
tr W2(u)duds ~ -1
t
Itto
tr W(s)ds
t --+
00
(7.2.72)
Itto tr W2(s)ds
2 (s )ds
--+
00
as
00.
as t --+
21tto
+M < t
00
and so
tr W(s)ds.
(7.2.73)
Now by the Cauchy-Schwarz inequality, we have
I~
10
t
tr W(s)ds
[1 10
10
~ t t(tr W(s))2ds ]1/2 ~ [

~ t tr W2(s)ds]1/2
so (7.2.73) gives
[ ~t Jtot Jtor tr W 2(U)dUdS]
< 4n
t
Jto
tr W 2(s)ds.
(7.2.74)

If we set H(t)
h: h: tr W2(u)duds,
H2(t) < 4ntH'(t),

and so
t::::: tl
519
then from (7.2.74), we obtain
for some large
tl::::: to
(7.2.75)
1 < H'(t)
4nt - H2(t) ,
(7.2.76)
Now an integration of (7.2.76) over [tl' 00) gives 00 = (l/4n)lntl:tl s::

1/H(tl) < 00, which is a contradiction. Thus, limt-+oo
tr W2(s)ds
h:
exists as a finite limit. This implies the existence of limt-+oo ftto W2(s)ds,
as the following argument shows: Let the (operator) norm of a matrix A be
denoted by IAI. For to s:: s s:: t, define A[s, tl by A[s, tl =
W 2(u)du.
Then, A[s, tl is a nonnegative definite matrix and
f:
IA[s, tll = Al(A[s, t])
s::
tr Als, tl = 1 \ r W2(u)du -+ 0 as s, t -+
00.
f:
Hence, we have IA[s, tll -+ 0 as s, t -+ 00, i.e.,

W2(u)du -+ 0
as s, t -+ 00, which implies the existence of limt-+oo It: W 2(s)ds. This
completes the proof of the lemma.
Now we are ready to prove the following two results.

Theorem 7.2.5. Assume that condition (7.2.66) holds. Then equation
(7.2.5) is oscillatory in case anyone of the following conditions is satisfied
(Id
lim sup
(12)
lim sup
t-+oo
(13)
(14)
t-+oo
~ 1.t Al (1.
t to
Q( u )du ) ds
~ 1: [AI (1: Q(U)du)f ds
limapproxsup Al
t -+
to
00
limapproxinf Al
t -+ 00
(1. Q(S)dS)
(1. Q(S)dS)
t
to
t
to
(7.2.77)
00,
00,
(7.2.78)
00,
(7.2.79)
- 00.
(7.2.80)
Proof. (II) Assume conditions (7.2.66) and (7.2.77) hold. Suppose there
exists a prepared solution X(t) of equation (7.2.5) which is not oscillatory.
Without loss of generality, we may suppose that det X(t) i- 0, t:::::: to so
that from equation (7.2.10), we have
(7.2.81 )
Chapter 7
520
By the convexity of ),1 and the fact that ftto W2(s)ds ~ 0, t ~ to, (i.e.,
nonnegative definite for t ~ to), we find from (7.2.81) that ),l(W(t)) +
),l(-W(tO))
lit
to
~),1 (It: Q(s)ds), and hence
),l(W(s))ds
t - to
+ --),l(-W(tO))
~t
1t (1
to
A1
to
so that from condition (7.2.77) there exists a sequence

as n ---+ 00 wi th
T1
iT
n to
A1(W(s))ds ---+
Q(U)dU) ds
{Tn}~=l'
Tn ---+
as
00
(7.2.82)
Tn ---+ 00
(7.2.83)
00.
Since AdW(s))2 S ),1(W2(s)) by the Cauchy-Schwarz inequality, we have
[L t
(),,(W(')))'d'l 'I'
" [:. J.>(W'('))df' -> 00
as Tn ---+
00.
But since L(Q) > -00, Lemma 6.2.3 implies that ft'; A1(W 2 (S))ds S
ft'; tr W2(s)ds < 00. This contradiction proves part (It).
(1 2 ) Assume conditions (7.2.66) and (7.2.78) hold. As in part (11) above,

we may assume W(t) = -X'(t)X-1(t) is a solution of equation (7.2.10)
for t ~ to, and so by Lemma 7.2.3, we obtain
W(t)
where 0 = -W(to) (A1(1: Q (S)ds))2
roo W2(s)ds + 0
it
Q(s)ds,
(7.2.84)
ft'; W 2(s)ds. Hence, we have

=
()'l[W(t)+ 1=W 2(S)ds+c]r
<
),1
(W(t)
< tr (W(t)
1
+1
1
+
00
00
< 2tr (W(t) +
< 4tr W 2(t) +4tr
W2(s)ds
W2(s)ds
00
0)
or
W 2(S)dS) 2 + 2tr 0 2
(1=
W 2 (S) dSr
+2tr 0 2,

(since tr (A
+ B)2
~ 2(tr
~ 1]Al (1: Q(U)dU)
521
A2 + tr B2. Therefore,
ds
~ ~ l>r
41t
+t
to
(1
00
W2(u)du
tr W (s)ds
ds
+ 2(t - to)
t
tr C .
(7.2.85)
If we set A(t) = ftOO W2(s)ds, then tr A(t) -+ 0 as t -+ 00 and A(t) ;::: 0
so that Al(A(t -+ 0, and hence Al(A2(t -+ 0 as t -+ 00. Therefore,
tr A2 (t) -+ 0 as t -+ 00. Thus, the integrals on the right-hand side of
(7.2.85) tend to zero as t -+ 00, and the last term is bounded. However,
condition (7.2.78) implies that the left-hand side of (7.2.85) is not bounded.
This contradiction completes the proof of part (12).
(13) Assume conditions (7.2.66) and (7.2.79) hold. As in part (1 2 ), we
obtain equation (7.2.84) so that
Al(W(t +
00
Al(W2(sds ;::: Al
(1:
Q(S)dS) + An(-C).
(7.2.86)
!too
Al(W 2(Sds -+ 0 as t -+ 00, it follows that ftoo [At{W(s)j2ds

Since
-+ 0 as t -+ 00 and Al (ftOO W 2(s)ds) -+ 0 as t -+ 00. Now for any
k ;::: 1, p, {t : Al (ltto Q(s)ds) ;::: k} =
00
so that if k;::: IAn( -C)I
+ 1,
then from (7.2.86), we have

(7.2.87)
(7.2.88)
i.e., fEJAl(W(S)2ds = 00, where Ek
contradiction proves part (13).
= {t : Al(W(t ;::: 1/2}. This
(14) Assume conditions (7.2.66) and (7.2.80) hold. The proof is similar to
part (13). Since for any M > 0,
(7.2.89)
and since
Chapter 7
522
it follows that if M> l+IAl(-C)I, we have p,{t: Al(W(t)) :::; -I} = CXJ
so that
(AI (W(s)))2ds = CXJ, which is a contradiction. This completes
the proof of the theorem.
It':
Theorem 7.2.6. Assume condition (7.2.66) holds. Then equa.tion (7.2.5)

is oscillatory if
An (
limapproxsup
t --+ 00
Proof. Assume L(Q) =
-CXJ,
limapproxsup
t --+ 00
An (
lto
Q(S)dS) > -
(7.2.90)
CXJ.
and
lto
Q(S)dS) = m > -
CXJ.
If equation (7.2.5) is nonoscillatory, then without loss of generality there

is a solution X(t) of (7.2.5) for which det X(t) =f. 0 on [to, CXJ).
Then with W(t) = -X'(t)X-l(t), t ~ to, we get equation (7.2.10). By
Lemma 7.2.3, since L(Q) = -CXJ, it follows that tr
W2(s)ds -+ CXJ
as t -+
Al ( -
CXJ,
and hence
Al
(Itto W
It:
2 ( S ) dS)
Itto Q(s)ds) = -An Utto Q(s)ds) ,
Al (W(t)
-1:
Q(S)dS)
-+
CXJ
Al(W(t)) - An
(-1:
(1:
{t : (It: Q(s)ds) ~ m -
1
-tr
n
it
to
W2(s)ds
-+
-1:
-1:
-1:
-1:
Q(S)dS)
(7.2.91)
Q(S)dS) .
E}
CXJ.
~tr
(W(t)
Q(s)ds - W(to))
~tr
(W(t)
Q(S)dS) -
< Al (W(t)
Since
CXJ.
from equation (7.2.10), we have
< Al(W(t)) + Al
Thus for any E> 0, p,

An
from equation (7.2.10), we have
as
Q(S)dS) -
< Al (W(t)) + Al (
Al (W(t)) - An
(1:
Now a.gain
~tr W(to)
~tr W(to)
Q(S)dS) -
Q(S)dS) -
~tr W(to)
~tr W(to)
(7.2.92)
523
and since (l/n)tr It: W2(s)ds ;::: (l/n) It: A1(W 2(Sds
follows that
~tr W(to) + ~
1:
A1(W2(sds ::::; A1(W(t - An
(1:
from (7.2.92) it
Q(S)dS) , t;::: to
(7.2.93)
Hence, for every E> 0,
{ l I l+ -t
Jl t: -tr W(to)
n
to
AdW2(sds ::::; A1(W(t-m+E} =
00
(7.2.94)
and since It: A1(W2(sds;::: A1 (It: W2(s)ds) ---+

that if E is defined by
00
as t ---+
00,
we find
then Jl(E) =
>'1 (W2(t
00. But now with P(t) = It: A1(W2(sds, we have P'(t) =

;::: A1(W(t2, and so P'(t);::: (1/4n 2)p2(t) for tEE, and
hence it follows that
r P'(s)
lE p2(s)ds
;::: 4n 2Jl(E)
00,
which is a contradiction since
r P'(s)
lE P2(S) ds
::::;
P(to
+ 1)'
To prove our next result, we need the following lemma.
Lemma 7.2.4. Let p(t) be locally bounded, nonnegative and measurable

on [to, (0) with p(t) not almost everywhere zero. Let q(t) be nonnegative
and locally integrable such that
lto
p(s)ds ;::: q(t)
lto
ls
p2(u)duds
for almost all t;::: to.
Then for all sufficiently large T;::: to, q(t)/0 E
Proof. Let P(t)
= It: It:p2(u)duds.
.c2[T, (0).
Then, P(t) is continuously differ-
entiable and P(t) > 0 for t > t 1, say. Clearly, P'(t)

by the Cauchy-Schwarz inequality
P'(t) ;:::
(7.2.95)
to
(_1_)
(it P(S)dS)
t - to
to
1/2
;:::
= It:p2(s)ds
(_1_)
q2(t)p2(t).
t - to
and
(7.2.96)
Chapter 7
524
Now, since I;'(P'(s)/p 2 (sds <
00
for T> t 1, the result follows.
Theorem 7.2.7. Assume that
Al
(1:
> 0 for sufficiently large t
Q(S)dS)
(7.2.97)
and
liminf
Hoo
Al (Itt Q(s)ds)
0
> O.
(7.2.98)
An (It: Q(s)ds)

lim
t->oo
~t itot Al (1to Q( U)dU) ds

8
(7.2.99)
00.
Proof. The argument is similar to that of used in Theorem 7.2.4, so we

content ourself here with sketching the proof. If (7.2.5) is nonoscillatory,
we may without loss of generality find a prepared solution X(t) with
det X(t) =F 0 on [to, 00). If Wet) = -X'(t)X-1(t), then Wet) is
symmetric and we have
cI>(t) = Wet) -
t W2(s)ds
ito
it
Q(s)ds + C,
i 2: to
(7.2.100)
to
for some constant matrix C.

As in the proof of Theorem 7.2.4, we obtain (7.2.39). Let U(t)
the orthogonal matrix whose rows are the ei(t) and let Ci(S, t) be
projection of yet) onto ei(s). Let the components of U- 1(t)c(to, t)
viet) and let the components of [U- 1(S)-U- 1(t)]c(to,00) be Wi(S, t),
1,2, ... ,n. Then, we have
Al
(1:
=
Q(S)dS)
+ y*(t)Cy(t)
tJ.li(t)vt(t)
i=l
-it
be
the
be
i =
y*(t)cI>(t)y(t)
J.lT(s) [Vi(t)
+ Wi(S, t)]2ds.
to i=l
Again as in the proof of Theorem 7.2.4 the given hypotheses of the theorem
allow us to find a unit vector yet), functions Pij(t), i = 1,"', n, j =
1, ,m and ~ > 0 such that
~Al
(1:
Q(S)dS) -
211CII < y*(t)cI>(t)y(t)
2: [Pi j (t) - ~
-,1
1:
(7.2.101)
Pij (s )dS] .
525
Hypothesis (7.2.99) and (7.2.101) imply that there exists a tI ;::: to such
that
(7.2.102)
If jlij == 0 a.e. for t;::: to, let ijij(t) = 0, otherwise define ijij by
(7.2.103)
By Lemma 7.2.4 each ijij has the property that ijij(t)IJi E 2[T, (0) if
> tI is sufficiently large, and so ij( t) = 2::i,j ijij (t) E 2 [T, (0) if T IS
sufficiently large. However, (7.2.102) implies that
[ij(t) -
l2:.1: 1:
fLij(u)duds > 0 for
t >T
',J
and so ij( t) >
for
t > T,
contradicting the fact that
2 [T, (0). This proves the theorem.
ij( t) / Vi E
To illustrate Theorems 7.2.5 and 7.2.7, we present the following counterexamples.

Example 7.2.3.
In Theorem 7.2.5 parts (h) and (h), one cannot re-
(It:
(It:
Q(s)ds) , or [>'1
Q(s)ds)
by these
place the average of >'1
expressions alone, and that one cannot replace 'lim approx sup' by 'lim
sup' in parts (13) and (14). For this, it suffices to consider a scalar counterexample. We choose v(t) E CI([to,oo),IR) n 2[to, (0) and define
q(t) = v'(t) - v 2(t). Then, x(t) = exp ( -
It: v(s)ds)
is a nonoscillatory
+ q(t)x(t) = 0, and we have It: q(s)ds =

v(t) - v(to) v 2 (s)ds. It is clear that we can choose v(t) such that
limsuPt-+oo v(t) takes on any value M with
- 00 < M :::; 00,
liminfHoo v(t) takes on any value m, -00:::; m :::; M :::; 00 and such
that liminfHoo(1/t) Itto v(s)ds > -00.
solution of the equation x"(t)
It:
Example 7.2.4.
The condition (7.2.99) is not by itself an oscillation
criterion for (7.2.5) when n> 1. To see this consider the case when Q(t)
is a 2 x 2 diagonal matrix for all t, say Q(t) = diag {q(t), q2(t)}. Then
(7.2.5) is uncoupled as two scalar equations
x~'(t)
+ qi(t)Xi(t) = 0,
= 1,2.
(7.2.104)i
If we take to = 0 and set Pi(t) = I~ qi(s)ds, we will have the required

counterexample if we can find qI (t), q2(t) so that equations (7.2.104h and
Chapter 7
526
(7.2.104h are both nonoscillatory with
P2(t) ;:::
and
lim -1
t--->oo
it
0
pi(s)ds
(7.2.105)
pi(t) = max{0,P1(t)}.
where
00,
t~
for
(7.2.106)
For this, on choosing
C1(t)
1, tEsupportpi(t)
and defining yet)
Al
0, otherwise,
= (C1(t),C2(t))*,
(lot Q(S)dS)
0, t E support pi(t)
1, otherwise
we have
(lot Q(S)dS) yet)
;:::
y*(t)
Ci(t)P1(t)
+ C~(t)P2(t)
;::: pi(t).
Thus condition (7.2.99) holds, but clearly equation (7.2.5) is nonoscillatory.

The choice of q2(t) is straightforward, e.g., q2(t) = 1/(4(1 + t)2).
To construct a suitable q1 (t), we first give an inductive definition of a
suitable Riccati scalar variable W1(t). Define W1(t) on [0,1] by W1(t) =
{ ~: !~ ~~ ~ t ::::;
so that W1(t) is continuously differentiable and
1,
10
monotone nondecreasing on [0,1]. Set a1 = 1 wi(s)ds, b1 = 4a1 and

f1 = 1/(8b1). Then, a1;::: 1/2, b1 ;::: 2 and 2f1 ::::; 1/8 < 1/2. Let
Sm = E:':11/i. Extend W1(t) to [Sl, S2] by taking W1(t) = b1 on
[1 + f1, S2] and making WI (t) continuously differentiable and monotone
nondecreasing on [0, S2]. Inductively, suppose that ai, b;, fi have been
defined for 1::::; i ::::; m with bi = 4ai, fi = 1/(8bi ) < 1/(2(i + 1)),
and WI (t) has been defined as a continuously differentiable monotone
llondecreasing function on [0, Sm+1] such that WI (t) = b1 on lSi +Ei, Si+1]
and I;' wi(s)ds = ai, 1::::; i ::::; m. Then,
am+l ;::: am
+I
m+2E~
w~(s)ds = am
+ b~fm >
S7n+m
By the inductive hypothesis, we find that
am+l ;::: ( ~) m a1 ;:::

If bm+1 = 4am+1, fm+1
( ~)m
2
= 1/(8bm+l),
~ (~) m
then in view of
(l+!)m > l+ m > 2(m+2)

2
16
2'am.

we have
527
(3)-m
1
<
- 2(m+2)"
Now we extend WI (t) to [Sm+l, Sm+2] so that WI (t) = bm+1 on [Sm+l +
E
m+
1
<
-16
-2
-
Em +l, Sm+2] and is continuously differentiable and monotone nondecreasing

on [0, Sm+2]. With this inductive definition of WI (t) on [0,(0), note
that if t E [8 m + em, Sm + 2Em], we have
1L'dt) -
iot
'IL,!
(s)ds ::0: bm
am -
2Embm ::0: "2bm.
Define ql(t) by ql(t) = w~(t) - wi(t). Then for t E [Sk + Ek,Sk + 2Ek],
we find Pl(t) =
ql(s)ds::O: (1/2)b k . Thus, if Sm :s; t:S; Sm+l, we obtain
Jd
I1t
t
Since
Sm rv
t1 L
m
pt(s)ds::O:
In m for sm:S; t
11t
k=1
:s; Sm+l, we have m
p+(s)ds
1
(7) .
"2 bkEk = 16
rv
-1
16
t)
( ~
rv
et so that
Thus (7.2.106) holds, yet (7.2,104h is nonoscillatory.

Finally, we turn our attention to variational principles for obtaining
oscillatory criteria for (7.2.5). We assume that Q(t) is symmetric and
locally integrable on [tOloo). On any subinterval [0,;3] of [to, (0), we
define
lR n : 7](0) = 7](;3) = 0,
7](t) E AC[o,;3], 7]'(t) E L 2(0,;3)}.
(7.2.107)
The basic result we need here is that system (7.2.5) is oscillatory if and
only if there is a sequence of intervals {[an, bn ]} with limn-+oo an = 00
and a sequence of functions {<Pn}, <Pn E Al (an, bn) such that
A 1 (oc,;3)
{T,: [0,;3]
(7.2.108)
The idea now is to utilize conditions on the behavior of Q(t), or its integral
to construct sequences fulfilling (7.2.108).
Our first result in this direction is the following.
limSUp.Al
t-+oo
(it
to
Q(8)dS)
00.
(7.2.109)
Chapter 7
528
Then equation (7.2.5) is oscillatory if either
(id AI(Q(t)) is bounded above on [to, 00), or
(h) An(Q(t)) is bounded below on [to, 00).
Proof. Choose any sequence ak
> to with limk->oo ak = 00. Since
by condition (7.2.109) it follows that

Therefore, we can choose bk
that
Yk
(l:
limsuPHoo Al
> ak+2 and Yk E
Q(S)dS) Yk 2: k,
lR n
(I:k Q(s)ds) =
with Ilxkll
k = 1,2, .. .
00.
= 1, so
(7.2.110)
Suppose that condition (h) holds. Then there exists M such that Al(Q(t))
~ M for t 2: to. Define (/>k E A(ak' bk) as follows:
(t - ak)Yk, ak ~ t ~ ak + 1
(/>k(t) = { Yk, ak + 1 ~ t ~ bk - 1
(-t + bk)Yk, bk - 1 ~ t ~ bk.
Clearly,
J:: 1<t{(s)1 ds = 2,
2
(7.2.111)
and
Thus, for sufficiently large k,
It follows that system (7.2.5) is oscillatory.
If condition (i2) holds, we have An(Q(t)) 2: -M for t 2: to. Define

>kEA1(ak-1,bk+1) by
>k(t) =
(t + 1 - ak)Yk, ak - 1 ~ t ~ ak
{ Yk, ak ~ t ~ bk
(-t + bk + l)Yk, bk ~ t ~ bk + 1.
(7.2.113)

Again J::~: 1{(sWds
= 2,
and we have
bk+l
l ak- 1>'k(s)Q(s)1>k(s)ds
1
1:~1 1>'k(s)Q(s)1>k(s)ds +
+
1:
529
1:d1
1>'k(s)Q(s)1>k(s)ds
1>'k(s)Q(s)1>ds)ds
k
k
[s + 1 - ak]2Y'kQ(S)Ykds
l akb Y'kQ(s)Ykds + la
ak-1
bk + 1
+
[b k + 1 - t]2Y'kQ(s)ykds
bk
k- M
[1:~1 [s + 1 -
ak]2ds
l:
+1
[bk
(7.2.114)
+ 1 - s]2ds
k--M
3
'
and so for sufficiently large k,
bk+l
l ak- [11>~(s)12
1
- 1>'k(s)Q(S)1>k(S)] ds <
2+
3M
- k < 0,
and again we conclude that equation (7.2.5) is oscillatory.
Example 7.2.5. In Theorem 7.2.8 condition (7.2.109) alone is sufficient

has been shown by Byers et. al. [12]. However, this condition cannot be
replaced by limHoo ft~ >'1 (Q(s))ds = 00. In fact, Example 7.2.4 can be
used to provide a counterexample. For this, we define ql(t), q2(t), Wl(t),
etc. as in Example 7.2.4, then
lot Al(Q(s))ds
lot q+(s)ds
lot[w~(s) -w~(s)]+ds.
Nowon [sm,sm+Em], we have wl(s)::;bm , andso
1~rn+<m[w~(S) -w~(s)]+ds ~ 1~n+<'1w~(s) -w~(s)]ds

=
Wl(Sm + Em) - Wl(Sm) -
j sm+<rn
w~(s)ds
Sm
Chapter 7
530
and hence,
Theorem 7.2.9. Let Q(t) = (qij(t)), i,j = 1"", n and each qij(t) is
bounded on [to,oo), and qii(t) = 0, i = 1"", n. Then system (7.2.5) is
oscillatory if there exist i, j such that
limsuplt qij(s)ds =
t-+oo
(7.2.115)
00.
to
Proof. Applying the technique of Theorem 7.2.8 to the scalar equation
x"(t)
+ Qij(t)x(t)
(7.2.116)
there exist intervals [am, bml with limm-+oo am =

= 1) such that
00
and a scalar function
[1~(s)12 - Qij(s)lm(s)1 2 ] ds < O.
(7.2.117)
m E AI(am,bm ) (with n
Ia~
b~
Define 'm E Al (am, bm ) by 'm(t) = col (0, ... , m(t), 0, ... , m(t), ... ,0),
where the only nonzero entries are the ith and jth. Then,
1~n~['11/(SW-,;"(S)Q(S)'(S)] ds = 21:n['~(SW-qij(s)"m(s),~ ds < 0

implies oscillation of (7.2.5).
Theorem 7.2.10. Suppose for each positive integer m::::: to there exists
a positive number Em, and for each positive integer k there exists a unit
vector Ymk E 1Rn such that the set
has measure at least Em. Then equation (7.2.5) is oscillatory.

Proof. Let m::::: to + 1 be a positive integer. Without loss of generality
we can assume that 0 2 + 12amEm.
(7.2.119)
531

Let
be the unit vector given by the hypothesis of the theorem. Since

Em, we can find a finite collection of closed intervals I j =
[aj, bj ], 1 ::; j ::; r where m < a1 < b1 < ... < a r < br are such that
Uj=lIj C Smk and p (Uj=lIj) = 2:;=1 8j ~ Em /2, where 8j = bj - aj.
Since
Q( s )ds is a continuous function of t, we can by maximizing
the length of the intervals, assume that
Ymk
p(Smk)
I!
(/~j Q(S)dS) Ymk
Y;"k
and
Y;"k
(I:j
Q(S)dS) Ymk
k,
1::; j ::; r
(7.2.120)
1::; j::; r-l.
(7.2.121)
k,
Furthermore, by discarding intervals if necessary and shrinking the final

interval, we can suppose that
(7.2.122)
Let ao and bo be defined to m, <50 to be 0, and let 6.j
= 0,1,, r.
= 2:{=0 <5j ,
We define rjJm E Adm - (l/2)E m, br ) as follows
rjJm(t)
(7.2.123)
Pm(t)Ymk,
where
Pm(t)
+ (1/2)Em - m, m - (1/2)Em ::; t ::; m

(1/2)Em - 6. j , bj ::; t::; aj+1, 0::; j ::; r - 1
(1/2)Em - 6. j - 1 + aj - t, aj::; t ::; bj+l, 0::; j ::; r.
(7.2.124)
Clearly,
1
2Em
and
l br
rjJ:n(s)Q(S)rjJm(s)ds
m-(1/2)f m
+~
j=O
+ 6.r
(7.2.125)
lbr
rjJ:n(s)Q(s)rjJm(s)ds
m-(1/2)f~
bj
i+
rjJ;"(s)Q(s)rjJ(s)ds
+~
j=l
bj
rjJ;"(s)Q(S)rjJm(s)ds.
aj
(7.2.126)
Chapter 7
532
By (7.2.120) and (7.2.121), it follows that
;"(s)Q(S)m(s)ds 2: -am
m-(1/2)f~
[s + (1/2)Em - m]2ds
m-(1/2)f~
-24 amEm
(7.2.127)
and by (7.2.120),
r ' ;"(s)Q(s)m(s)ds
Jb o
2Emk .
(7.2.128)
For 1::; j ::; r - 1, by (7.2.120) and (7.2.121) we also have
aH1
;"(s)Q(s)</>m(s)ds = ((1/2)Em - .0.j )
((1/2)Em - .0.j)
</>;"(s)Q(s)</>m(s)ds
l bj
p;,,(bj
aH1
Y;'kQ(s)Ymkds
I:j Y;'kQ(s)Ymk ds - I: jY;'kQ(S)Ymk ds
Noting that for 1::; j ::; r, Pm(t) 2: 0 and

hence on integration by parts, we find
l bj
l bj
= -Ion
O.
(7.2.129)
and
[aj, bj ],
p;"(S)Y;"kQ(S)ymkds
Y;"kQ(s)Ymkds
p~(t)
l bj
2Pm(s)
(is
Y;"kQ(U)Ymkdu) ds.
(7.2.130)
By (7.2.120) and (7.2.121), we further have
l bj
Y;"kQ(S)Ymk ds
I:
Y;"kQ(s)ymkds -
I:
y;'kQ(S)ymkds 2: 0,
(7.2.131)
(with equality if 1 ::; j ::; r - 1). Since t E [aj, bj ] implies t E Smk,
using (7.2.120) we find that for t E raj, bj ],
From (7.2.130) - (7.2.132), we get
l bj
J
</>;"(s)Q(s)</>m(s)ds 2: 0,
= 1,, r.
(7.2.133)
533
Finally, from (7.2.125) - (7.2.129), (7.2.133) and (7.2.119), we find
m-/2)
[11>'(s)1 2 -1>;"(s)Q(S)1>rn(S)] ds s:
Ern
+ 2amE~ 4
2Emk < O.
Since m - (1/2)Em > m - (1/2) is an arbitrary positive integer exceeding

to, the theorem follows.
7.3. Oscillation Theorems for Differential

Systems with Functionally
Commutative Matrix Coefficients
Consider the second order nonlinear differential system
X"(t)
+ G(X*(t))Q(t)H(X(t))
= 0
(7.3.1)
where
(i) G, HE F, and the set F will be defined later,
(ii) Q(t) is a real-valued, continuous, symmetric n x n matrix valued
function on [to,oo), to > 0,
(iii) X(t) is a twice continuously differentiable n x n matrix valued
function on [to,oo), and X* denotes the transpose of X.
A solution of the system (7.3.1) is a real-valued n x n matrix X(t)
satisfying (7.3.1) on [T,oo), T 2:: to (T may depend on X) and not
identically singular on a neighborhood of infinity. Any solution X(t)
of (7.3.1) satisfies (7.2.7), and if it satisfies (7.2.6), X(t) is said to be
prepared. X(t) oscillates if del X(t) has arbitrarily large zeros.
The continuous n x n matrix-valued function
functionally commutative on an interval J if
Q(s)Q(t)
Q(t)Q(s)
for all
Q(t)
s, t E J.
is said to be
(7.3.2)
We shall consider the general system (7.3.1) with symmetric, functionally commutative coefficients and show that the oscillation theory for such
systems can be effectively reduced to the study of a diagonal system of
scalar second order equations. In the case of equations like (7.3.1), this
system of scalar equations is uncoupled, allowing one to directly employ
the existing theory to obtain oscillation and nonoscillation criteria.
The following notation and preliminaries are required: Let Mn be the
Banach space of real n x n matrices with any of the usual norms. Let Vn
and N n be the subsets of diagonal and nonsingular matrices, respectively.
534
Chapter 7
We denote by F the set of continuous functions F: Mn -* Mn which are

invariant on Vn and Vn nNn , and for which F(p-l DP) = p-l F(D)P
for all D E Vn and PENn. Let Sn denote the symmetric n x n
matrices. Evidently the values of F on Vn completely determine it on
Sn. Also the following properties hold:
(al) F is invariant of N n ,
(a2) F(A*) = (F(A))* for all A E Mn, in particular F is invariant on
Sn
Let F(diag{xl"",xn}) = diag{h(xl,,Xn),-,fn(Xl,,Xn)},
Denote by Fo the subset of F for which !i(Xl,"',Xn ) = !(Xi), i =
1,2"" ,no
As examples of F E Fo, we can take F(X) = X k where k is a
nonzero integer, or F(X) = X1/q where q is an odd integer.
It is easy to check that both F and Fo are closed under multiplication.
In what follows, we shall consider the linear system (7.2.5) and the
general system (7.3.1), where G, HE F, and Q is continuous, symmetric
and functionally commutative. We shall denote the set of such Q by C.
The following lemma gives a characterization of functionally commutative matrices.
Lemma 7.3.1. Let the n x n matrix A(t) be diagonalizable, and
functionally commutative on an interval J. Then, A(t) has a constant set
of eigenvalues, and therefore there exists a constant matrix P such that
A(t)
p- 1 D(t)P
where
D(t) E V n ,
t E
J.
In the case of linear system (7.2.5), we can readily reduce our discussion
to that of an associated diagonal system.
Theorem 7.3.1. Let Q E C, and assume that there exists a constant matrix P so that Q(t) = P-lD(t)P, where D(t) = diag{ql(t),,qn(t)},
where qi(t) E C([to, CXl), lR), i = 1,2, ... , n. Then all prepared solutions
of (7.2.5) oscillate if and only if at least one of an associated system of n
scalar equations
yr(t) + qi(t)Yi(t)
0,
i = 1,2, ... , n
(7.3.3)
is oscillatory.
Proof. Since Q(t) is symmetric, the matrix P is actually orthogonal.
Let yet) = PX(t), then the system (7.2.5) is transformed into
Y"(t)
+ D(t)Y(t)
O.
(7.3.4)
535
Now X(t) is a prepared solution of (7.2.5) if and only if Y(t) is a prepared

solution of (7.3.4) and evidently both systems have the same oscillatory and
nonoscillatory character. To establish the oscillatory character of (7.3.4),
it is enough to consider the solution Yo(t) satisfying
Yo (to)
= 0,
Y~ (to)
= I,
is the identify matrix.
(7.3.5)
In fact, in view of the Morse separation theorem [42] the oscillation of this
solution will guarantee oscillation of all solutions. Clearly, the solution of
(7.3.3) satisfying the initial condition (7.3.5) is diagonal for all t:::: to,
and so for this solution, system (7.3.4) may be written in the form (7.3.3),
where Yo(t) = diag{vdt),"',Vn(t)}, cmd D(t) = {lJl(t),,qn(t)}.
Now since det Yo(t) = TI~=l Yi(t), Yo(t) oscillates if and only if one of
the functions Yi(t) oscillates, i.e., if and ouly if one of the scalar equations
(7.3.3) is oscillatory. This completes the proof.
Now we shall study the general system (7.3.1). For this, first we note the
following simple consequence of a solution which is symmetric and prepared.
Theorem 7.3.2. Any symmetric and prepared solution X(t) of (7.3.1)
which is nonsingulaT on an interval J, commutes with Q(t) for all t E .J.
Proof. Since X(t) is symmetric and prepared, (7.3.1) shows that X(t)
commutes with X'(t) for all t:::: to. Differentiation of the identity
X'(t)X(t) - X(t)X'(t) = 0
leads to the fact that X(t) commutes with X"(t) for all t:::: to. If X(t)
is nonsingular on J, system (7.3.1) gives X(t)G(X(t))Q(t)H(X(t)) =
G(X(t))Q(t)H(X(t))X(t). Since H(X(t)) is nonsingular on J, we can
wri te this as
X(t)G(X(t))Q(t) = G(X(t))Q(t)H(X(t))X(t)H-l(X(t))
t E J.
(7.3.6)
Now, let X(t) = P-1(t)Y(t)P(t), where P(t) EN,., and X(t) E Dn.
Then, since Y(t) and H(Y(t)) E Dnl we have
H(X(t))X(t)
P-l(t)H(Y(t))Y(t)P(t)
for all
p- 1(t)Y( t)H(Y( t) )P( t)
X(t)H(x(t)).
Similarly, X(t)G(X(t))
= G(X(t))X(t). Thus, (7.3.6) becomes G(X(t))

G(X(t))Q(t)X(t) for all t E J. The result now follows from
the fact that G(X(t)) is nonsingular on J.
xX(t)Q(t)
Next, we state the following two well known lemmas.

Lemma 7.3.2. Let A be an n x n diagonalizable matrix. Then the
matrix S diagonalizes A, i.e., SAS- 1 E Dn if and only if S is a matrix
whose columns are linearly independent eigenvectors of A.
536
Chapter 7
LeIIlIIla 7.3.3. Let AI,"', AN be a given set of n x n matrices. Then

the following statements are equivalent:
(I)
A commute pairwise.
Ai is diagonalizable for each i, and the
(II) AI,"', AN can be simultaneously diagona.lized.

TheoreIIl 7.3.3. Let Q E C be as in Theorem 7.3.1. Moreover, assume that G, HE:F and let F = GH with F(diag {Xl,""X n }) =
diag {II (Xl, ... , Xn), ... , !n(XI, ... , x n )}. Then all symmetric, prepared
solutions of (7.3.1) oscillate if and only if the diagonal system
y~/(t)+qi(t)h(YI(t)'''',Yn(t
i=1,2, .. ,n
0,
is oscillatory, i.e., if and only if in each solution (Yl (t), ... , Yn (t

(7.3.7) at least one of the scalar functions Yi(t) oscillates.
(7.3.7)
of system
Proof. Assume that (7.3.7) is oscillatory and (7.3.1) has a nonoscillatory

solution X(t). Then X(t) is nonsingular on [to, 00), (say). By Theorem
7.3.2, X(t) commutes with Q(t) for all t::;> to. X(t) is symmetric and
therefore diagonalizable, and so by Lemma 7.3.3 it can be simultaneously
diagonalized with Q(t) for t::::: to. By Lemma 7.3.2 it follows that for
each t::;> to, X(t) and Q(t) have the same set of linearly independent
eigenvectors. But Q(t) has a constant set of eigenvectors for t::;> to, so
X(t) and Q(t) can be diagonalized by the same constant matrix P for
all t::;> to. Now observe that
G(X(tQ(t)
G(p-lY(t)p)p-1 D(t)P,
P-IG(Y(tpp-1 D(t)P
yet) E Vn
P-IG(Y(tD(t)P
where
=
p- D(t)G(Y(tP = Q(t)G(X(t,
l
(7.3.8)
since G(Y(t, D(t) E V n .
Therefore, equation (7.3.1) can be written as
X"(t)
+ Q(t)F(X(t
0,
t::;> to.
(7.3.9)
Thus, we have
pylI(t)p-1
+ PQ(t)F(P-IY(t)p)P- 1
which reduces to
ylI(t)
+ D(t)F(Y(t =
O.
0,
(7.3.10)
Let yet) = diag {Yl(t),, Yn(t)}. Then, yet) is a nonsingular solution

of (7.3.10) for t::::: to. But (7.3.10) is equivalent to (7.3.7), and we have a
contradiction to the assumption that (7.3.7) is oscillatory. Therefore, the
system (7.3.1) is oscillatory.
537
Conversely, suppose that (7.3.7) has a nonoscillatory solution (YI(t),

,Yn(t)). Then, X(t) = P-IY(t)P = diag {YI(t), ,Yn(t)} is a
nonoscillatory solution of the system (7.3.1). This completes the proof.

Corollary 7.3.1. Let Q E C be as in Theorem 7.3.1 and G, HE Fo.
Let F=GH with F(diag{xI,,Xn}) = diag{J(xI),,f(x n )}. Then
all symmetric prepared solutions of the system (7.3.1) oscillate if and only
if all solutions (for some i), of
(7.3.11)
oscillate.
Next we shall present oscillation and nonoscillation theorems for the systems (7.2.5) and (7.3.1). First we shall consider the linear system (7.2.5).
Since in view of Theorem 7.3.1 oscillation of (7.2.5) is equivalent to oscillation of the uncoupled system of scalar equations (7.3.3), explicit knowledge
of the functions qi(t) (which exists if Q(t) is known explicitly) will allow us to use any of the known oscillation or nonoscillation criteria for the
second order linear differential equations of the form
y"(t)
+ q(t)y(t)
O.
(7.3.12)
As an application of Corollary 7.3.1 we present the following result.

Theorem 7.3.4. Let Q E C be as in Theorem 7.3.1 and suppose that
lim Al (
t-'>oo
Jtort Q( s )dS)
00.
Then all prepared solutions of the system (7.2.5) oscillate.

Proof. Since P-IQ(t)P = diag {qI(t),, qn(t)}, we find
so
Al (
(It:
t
Jt.rto Q(S)dS)
max
l:Si:Sn
Jtort qi(s)ds.
Itto
Now, limHoo Al
Q(s)ds) = 00 implies lim approx SUPHoo
qi(s)ds
for at least one index i. By Corollary 7.3.1 it follows that for such
indices i the equation y?( t) + qi (t )Yi (t) = 0 is oscillatory. The oscillation
of (7.2.5) now follows from Theorem 7.3.1.
= 00
538
Chapter 7
Remark 7.3.1. It is evident that weaker hypothesis

limapproxsup Al
t -> 00
(1 Q(S)dS)
t
to
00
would suffice. Other oscillation criteria similar to those in Chapter 2 and

Section 6.1 can also be applied to equation (7.2.5) when Q E C.
We conclude our discussion of the linear system (7.2.5) with the following nonoscillation result.
Theorem 7.3.5. Let Q E C be as in Theorem 7.3.1 and suppose that
lim
t--'>oo
1t
to
SAI(Q(s))ds
exists as a finite limit.
Then the system (7.2.5) is nonoscillatory.
It:
Proof. Let q(t) = AI(Q(t)). Then, limHoo

sq(s)ds exists as a
finite limit, which implies that equation (7.3.12) is nonoscillatory. Since
qi(t) :::; q(t) for all t 2: to, i = 1,2,,n it follows from the Sturm
comparison theorem that each of the equations (7.3.5) is nonoscillatory.
Hence the system (7.2.5) is nonoscillatory.

Next, we shall consider the nonlinear system (7.3.1) when Q(t) is
nonnegative definite for all t 2: to and G, H E:Fo with F = GH
and F(diag{ Xl, ... , xn}) = diag{f(xt}, ... ,f(xn )}. We recall the following
known oscillation criteria for the equation
x"(t) + q(t)f(x(t))
0,
where q(t) E C([to, 00), JR+) and f E C(JR, JR), xf(x) > 0,
for X =I O.
Joo du/ f(u) < 00.

Joo sq(s)ds = 00.
(II) Let
only if
(1 2 ) Let f(x)
(7.3.13)
f'(x) 2: 0
Then equation (7.3.13) is oscillatory if and
= xa/f3, where 0:, /3 are odd integers with 0 <, = 0://3 <
Joo siq(s)ds = 00.
1. Then equation (7.3.13) is oscillatory if and only if
We shall extend the above criteria to the system (7.3.1).

Theorem 7.3.6. Let the function f be strictly increasing with f(O) =
0, limx--,>oo f(x) = oo and suppose that
OO du
f(u) <
00.
(7.3.14)
539
Let Q E C be as in Theorem 7.3.1 and Q(t) nonnegative definite for

all t ~ to. Then all symmetric prepared solutions of the system (7.3.1)
oscillate if and only if
lim A1
t->oo
Proof.
limt->oo >'1
Itto sqi(s)ds =
00,
(it SQ(S)dS)
to
(I/o sQ(s)ds)
if and only if
= 00
I:7=1
00.
(7.3.15)
if and only if limt->oo max1::;i::;n
It: sqi(s)ds = 00,
It o
and since the
qi'S are all nonnegative if and only if limt->oo

Sqi (s )ds = 00, for at
least one index i. Theorem 7.3.6 now follows from the above criterion (1 1)
and Corollary 7.3.1.
Theorem 7.3.7. Let f(x) = xo./ f3 , where a, /3 are odd integers with
o oo
(it S'YQ(S)dS)
to
00.
Proof. The proof is the same as that of Theorem 7.3.6 except now we use
criterion (h).
Theorem 7.3.6 when applied to system
X"(t)
+ (X*(t))k
Q(t)Xk+l(t) = 0,
(7.3.16)
where k is a positive integer, and Q and X satisfy (ii) and (iii) of the
system (7.3.1) gives the following corollary.
Corollary 7.3.2. Let Q E C be as in Theorem 7.3.1 and Q(t) is

nonnegative definite for all t ~ to. Then condition (7.3.15) is necessary
and sufficient for all symmetric prepared solutions of the system (7.3.16) to
be oscillatory.
Finally, we present the following nonoscillation result. The arguments
are similar to those of earlier theorems and hence the proof is omitted.
Theorem 7.3.8. Let f(x) = X 2k +1, where k is a positive integer. Let

the eigenvalues of Q(t) be J.L1(t) ~ J.L2(t) ~ ... ~ J.Ln(t) ~ 0, and Q E C
be as in Theorem 7.3.1. Further, let J.Lj(t):::; 0 for each j, and
it
lim
s2k+lJ.L1(S)ds <
t->oo to
00.
Chapter 7
540
Then all symmetric prepared solutions of the system (7.3.1) are nonoscillatory.
7.4. Comparison Theorems for OperatorValued Linear Differential Equations

The Hille~Wintner theorem asserts that if q(t), ql (t) E C([to, (0), IR),
p(t) = ft q(s)ds, PI(t) = !too ql(s)ds, t 2: to exist and satisfy IPI(t)l:::;
p(t) for all t E [to, 00), then the existence of a nonoscillatory solution of
the equation x"(t) + q(t)x(t) = 0 on [to, (0) implies the existence of a
nonoscillatory solution of x"(t) +ql(t)X(t) = 0 on [to, 00).
OO
We shall consider the systems
X"(t)
and
+ Q(t)X(t)
X"(t) + QI(t)X(t)
(7.4.1)
(7.4.2)
and prove several generalizations of the Hille~Wintner result in the Banach

lattice case. To obtain such results we need the following notation, definitions and preliminaries. We denote by B, a Banach lattice. We recall
that a Banach lattice B is a Banach space with a vector lattice structure
such that Ixl:::; Iyl implies Ilxll:::; IIYII, x, y E B, where Ixl == xv( -x).
Let B+ = {x E B : x 2: O}. A subset S c B is order bounded if there
exists z E B+ with Ixl:::; z for all xES. The Banach lattice B is
said to be order complete if for every nonempty majorized (with respect to
ordering) subset A of B, sup A exists in B. Also, B is said to have
order continuous norm if each downward directed family A of B such
that inf A = 0 converges in norm to o. A Banach lattice with order continuous norm is automatically order complete ([46], Theorem 11.5.10, p.89).
Familiar examples of such Banach lattice are p, .c p, 1:::; P < 00, Co and
any reflexive Banach lattice.
Let .c+(B) denote the set of positive bounded linear operators on B,
i.e., T E .c+(B) if and only if T(B+) c B+ . .c(B) represents the
Banach algebra of bounded linear operators T: B -+ B. An operator
T E .c(B) is said to be regular if T = T I -T2 for some T I , T2 E .c+(B) .
.cr(B) denotes the vector space of regular operators on B. It can be
shown (d. [46]' p. 229) that if B is order complete, then T is regular
if and only if T maps order bounded sets onto order bounded sets if and
only if ITI == Tv ( - T) exists in the lattice structure induced on .c( B).
Furthermore, if B is order complete, then .cr(B) is an order complete
vector lattice ([46], p.229). We refer to [45) and [46) for further discussion
of Banach lattices and their properties.
541
Remark 7.4.1. If B is a Hilbert space, then (B) is a B"-algebra. In

this case comparison and oscillation theorems have been obtained which are
generalizations of the known theorems in the classical scalar case, (e.g., the
Sturm comparison theorem and the Hille-Wintner comparison theorem). In
these results the notion of comparison of operator in (B) is used which
is induced by the inner product, i.e., if T l , T2 E (B), then Tl 2: T2
means -< (Tl - T2)x, x >--2: 0 for all x in the Hilbert space B.
In this section we shall be interested in employing an alternative notion
of positivity of (B) which is induced by a vector lattice structure on
B. This has the advantage that the class of positive operators is not only
a positive cone but is also closed under multiplication, a fact which is not
true in the B"-algebra. case. A further advantage of the Banach lattice is
that it enables us to obtain certain results for nonself-adjoint equations.
Now, we consider systems (7.4.1) and (7.4.2), where Q, Ql : [to, 00) ---+
(B). By a solution X(t) of (7.4.1) (or (7.4.2)), we mean X: [to, 00) ---+
(B) which is twice continuously differentiable in the uniform operator
topology and satisfy the system (7.4.1) (or (7.4.2)) for t E [to, 00). A
solution X = X(t) is said to be nonsingular at a point tl E [to, 00) if
it has a bounded inverse X-l(t l ) E (B). If X(t) is nonsingular for all
t E [h, 00), tl 2: to then X = X(t) is said to be a nonoscillatory solution
of (7.4.1) on [tl' 00). Otherwise, X = X(t) is said to be oscillatory on
[to, 00). Note that the inverse X-l(t) of a nonoscillatory solution X(t)
of (7.4.1) is continuously differentiable.
We are now prepared to prove the following result.
Theorem 7.4.1. Let B be a Banach lattice with order continuous norm.

Q(s)ds, Pl(t) = limT ..... = Ql(s)ds
Suppose the limits P(t) = limT ..... =
exist (in the uniform operator topology of (B)), and
It
(i)
P(t), Pl(t), P(t) - Pl(t)
+(B), t
It
[to,oo).
If there exists a nonoscillatory solution X(t) of (7.4.1) such that

(ii) W(t)
= X'(t)X-l(t)
+(B) for all t
[to,oo),
then the system (7.4.2) has a nonoscillatory solution on [to, 00).
Proof. In what follows inequality sign will refer to an appropriate positive

cone. The Riccati transformation W(t) = X'(t)X-l(t) in (7.4.1) yields
the system
W(t) = W(T) +
iT
Q(s)ds +
iT
W2(s)ds,
We assert that (7.4.3) implies that the set
to
~ t <T
< 00. (7.4.3)
{It W 2(s)ds : to ~ T < oo}
is
an upward directed set in C(B) and is order bounded above by W(t). For
Chapter 7
542
{It W2(s)ds : to :S T < oo}
this, note that the mono tonicity of
follows
from the hypothesis that W(s) (and therefore W 2(s is nonnegative

W2(s)ds, so that (7.4.3) can be
for all s 2:: to. Denote by if>(t, T) =
written as
if>(t, T) = W(t) - W(T) - P(t) + P(T).
(7.4.4)
It
If to :S T :S T, then in view of (7.4.4), we have if>(t, T) :S if>(t, T) :S

W(t) + P(T), and hence
W(t)
+ P(T)
- if>(t, T) 2:: 0
for
to:S T :S T.
(7.4.5)
For each T 2:: T the left-hand side of (7.4.5) is in C(B), which is a normed
vector lattice and therefore has norm-closed positive cone .c+(B). Letting
T --* 00 in (7.4.5), we obtain W(t) - if>(t, T) 2:: 0,
which completes the
proof of the assertion. Since .cr(B) is order complete, it follows that
if>(t) = SUPT if>(t, T) exists. Let y E B+. Then, (7.4.3) implies that
W(T)y = W(t)y -
(iT
Q(S)dS) y - if>(t, T)y.
(7.4.6)
Since B has order continuous norm, if>(t, T)y --* if>(t)y in norm, so if>(t)
is the strong operator limit of if>(t, T) as T --* 00. Thus the limit in norm
as T --* 00 exists for the right-hand side of (7.4.6). We conclude that the
limit in norm of W(T)y as T --* 00 exists (in B+ by norm-closedness
of B+), we call it C(y). Thus,
C(y) = W(t)y - P(t)y - if>(t)y.
(7.4.7)
It follows that C(y) in fact defines a positive linear operator on B+ and

extends uniquely to a linear operator on B (cf. [46], p. 58). Thus (7.4.7)
holds for all y and we may write C(y) = Cy. Hence,
W(t) = C + P(t)
+ if>(t),
C E .c+(B),
t E [to, 00).
(7.4.8)
We now define a sequence of operator-valued functions {Zn (t)} as follows:
Zo(t)
Zn+l(t)
C, t 2:: to
=
+ P1(t) + s-limT .... =
iT Z~(s)ds,
n = 0, 1,, t 2:: to
(7.4.9; n)
where s-lim denotes the strong operator limit. To justify this, inductively
we shall show
O:S Zn(t) :S W(t) and Zn(t) is continuous in

the uniform operator topology .c(B), t 2:: to.
(7.4.10; n)
543
Clearly (7.4.10;0) is true. Now assume that (7.4.1O;n) holds. Then, 0:::;
Zn(s) :::; W(s) implies 0:::; Z;(s) :::; W 2(s). This is a consequence of
the cone + (B) being closed under multiplication. (It does not, in general, hold in the B*-algebra case). Thus,
{itT Z~(s)ds}
and is order bounded by W(t)
increasing in T
It W2(s)ds :::; W(t),
since
ItT
is monotone
1;1' Z~(s)ds :::;
and hence S-lilllT~=

Z~(s)ds exists. Therefore,
Zn+l(t) is well-defined by (7.4.9;n). Now it follows from the hypothesis
that 0:::; P 1 (t) :::; P(t) and 0:::; Zn+dt) :::; W(t). We also have
Zn+l(t) - Zn+l(S)
Pl(t) - P1 (s)
-1t Z~(T)dT,
(7.4.11)
so it follows that Zn+l(t) is continuous, hence (7.4.1O;n+1) is true and

thus (7.4.10;n) holds for all n 2: O. Let wn(t, T) =
Z;,(s)ds and denote
s-lim1'-+oo wn(t, T) = Wn(t). Since for fixed t, {Zn(t)}~=l is a monotone
increasing sequence in + (B), bounded above by W(t), it follows that
s-lirnn-+oo Zn(t) exists. Let Z(t) = s-limn-+oo Zn(t). By (7.4.9;n), we
have
(7.4.12)
Z(t) = C + H(t) + w(t),
ItT
where W(t)
s-limn-+CXl wn(t). Note that Z(t) is continuous, since
Z(t) - Z(s) = P1 (t) - Pl(S) -
s-limn~= it Z~(T)dT
(7.4.13)
and 0:::; Zn(T) :::; W(T) implies
o :::;
lilt
Z~(u)dull
:::; it [[W2(u)[[du --+ 0
as
t-
--+ 0
by continuity of W. For each T and each y E B+, Z~ (T)y --+ Z2 (T)y

in norm and Z2(T)Y is continuous in T. Now it follows from Dini's
theorem that this convergence is uniform on any compact subinterval [8, t]
of [to, (0). Thus, lim n -+ CXl
that s-limn-+oo
as
I:
Z~(T)dT
(Ist Z~(T)dT) Y = (Ist Z2(T)dT) y,

= 1:
Y E B+, so
Z2(T)dT. We may therefore write (7.4.13)

(7.4.14)
Dividing both sides of (7.4.14) by t - 8 (t i=- s) and taking limits in the

uniform operator topology as s --+ t, we get
(7.4.15)
Now define X(t) to be the solution of X'(t) = Z(t)X(t), X(to) = I
where I is the identity operator in (B). Then, X(t) is a nonoscillatory
solution of (7.4.2), and this proves the theorem.
544
Chapter 7
Example 7.4.1.
Let B be a Banach lattice with order continuous
norm and let Bl be its dual space. (Bl is also a Banach lattice, under
the dual norm, which is order complete). Let c E B+ and E Bt.
Let q(t) E C2([to,00),IR+) such that q'(t) > 0 and q"(t) < 0 for all
t E [to, 00). Define Q: [to, 00) ---+ .c(B) by
Q( t)y
q"(t)(y)c
1 + q(t)(c) ,
yE
Then,
Q(t) E .c+(B)
for all
t E [to, 00),
(jjQ(t)jj
jq"(t)jjjjjjjcjj )
1 + q(t)II(c)11 .
Note that
111
00
Q(s)dSII =
(1=
-1
+q~~:~(C)dS) 11llltcll ~
q'(t)IIllllcll
It may be verified that X"(t) + Q(t)X(t) = 0 has a solution X(t)

defined by X(t)y = y + q(t)(y)c, Y E B. Now it follows that X(t)
is a nonoscillatory solution and furthermore, the Riccati variable Wet) =
X'(t)X-l(t) is given by
Wetly
q'(t)(y)c
1 + q(t)(c) '
YE B
so Wet) E .c+(B) on (to, 00). Therefore, if Cl E B+, Cl ~ C, l E

Bt, 1 ~ and if ql(t) is a real-valued function such that
o -<
00
q~(s)
ds <
1 + ql(S)l(Cl)
-
00
q"(s)
1 + q(s)(c)
ds
t E [to,oo)
then equation X"(t) + Ql(t)X(t) = 0 has a nonoscillatory solution by

Theorem 7.4.1, where Ql(t) is given by
We note that in general Ql(t) rj. .c+(B).

One would like to weaken the hypothesis (i) in Theorem 7.4.1 to
(i)'
Pl(t)
.cr(B), pet), pet) -jPl(t)j E .c+(B), t E [to,oo)
and to dispense with hypothesis (ii) (which holds automatically in the scalar
case for a nonoscillatory solution X(t) of (7.4.1) when P(t);::: 0).
As far as (ii) is concerned, we may make the following remarks:
545

Assume P(t) E .c+(B), t E [to, 00), and define
Po(t) = P(t)
Pn(t) = P(t)
+ s-limT->~
.iT pLl (s)ds,
n = 1,2,.
Then the existence of nonoscillatory solution X(t) of (7.4.1) satisfying (ii)

is equivalent to the following:
(iii) Pn(t) is defined for all t E [to,oo), n
exists.
= 0,1,2"" and s-limn --+ oo Pn(t)
= P(t)
The equivalence of (ii) and (iii) can be established by using essentially the
same arguments as were used in the proof of Theorem 7.4.l.
When B is finite dimensional we can replace (i) by (i)'.
Theorem 7.4.2. Let B
P(t)
lim
T--+oo
iT
t
= lR n and suppose that
Q(s)ds,
(7.4.16)
exist such that (i)' holds. If there exists a nonoscillatory solution X(t)
of (7.4.1) such that (ii) holds, then the system (7.4.2) has a nonoscillatory
solution on [to, 00).
Proof. As in the proof of Theorem 7.4.1, we put W(t) = X'(t)X-l(t)
in (7.4.1) and obtain (7.4.3) which leads to (7.4.8). Let Z be the set of
continuous functions Z: [to, 00) --+ .c(lRn) such that IZ(t)1 ~ W(t) for all
t E [to, 00). Then, Z is closed, convex subset of M, the Frechet space of
continuous operator-valued functions on [to, 00) with the compact-open
topology.
Now if Z E Z, then for each t E [to,oo), Z(t) may be represented
by an n x n matrix, say, Z(t) = [(Z(t))ij]. Similarly, we may represent
W(t) by the n x n matrix W(t) = [(W(t))i'J, and we have I(Z(t))ijl ~
(W(t))ij, i,j = 1, .. , n. It follows that IZd(t)1 ~ W 2 (t). Furthermore,
we have IIZ2(t)11 ~ IIW 2(t)ll, where II II denotes the uniform operator
norm on .c(B). Since W 2 (t) is integrable on [to, 00), it follows that
Z2 (t) is integrable on [to, 00 ) , and
(7.4.17)
It follows from (7.4.8), (7.4.17) and (i)' that F on Z defined by
(7.4.18)
546
Chapter 7
is a map from Z into itself. Let {Zn}~=l' Z E Z with lim n - Hxl Zn

For any T 2: t > to, we find
= Z.
I IX) Z;(s)ds -IX) Z2(s)dsll

~ iT IIZ~(s) - Z2(s)llds + II1, Z~(S)dSII + II1, Z2(S)dsll
II(FZn)(t) - (FZ)(t)11
CXJ
CXJ
(7.4.19)
~
W2(s)dsll and
Z2(s) ds ll ~
W 2(s)dsll, we can for any given E > 0 choose T = T(E) so that
Z~(.s)dsll < c/3, n = 1,2,'"
and
Z2(s)dsll < E/3. By the
uniform convergence of Zn to Z on [to, (0), we have
Now since
III;
III;
III; Z~(s)dsil
III;
III;
III;
iT IIZ~(s)
- Z2(s)llds <
for all
n 2: N(E),
say
and all t E [to,T]. Thus, (7.4.19) gives II(FZn)(t) - (FZ)(t)11 <

all n 2: N(E), t E [to, (0). It follows that limn--+CXJFZn = FZ,
F : Z -+ Z is continuous.
for
i.e.,
Now for s, t 2: to, we have IIFZ(t) - FZ(s)11 ~ Ist IIQ1(r)lldr+

r) Iidr. Hence, F(Z) consists of equicontinuous, uniformly bounded
operator-valued functions. Thus, F(Z) is precompact. By Tychonov's
theorem, F has a fixed point in Z, i.e., there exists Z(t) such that
I: IIW2(
(7.4.20)
It follows that the system (7.4.2) has a nonoscillatory solution.
Next, we compare the system (7.4.2) in (JR. n )

scalar equation
xl/(t) + q(t)x(t) = 0,
with an appropriate
(7.4.21)
where q(t) E C([to,oo),JR.).

Theorem 7.4.3.
Let Ql : [to, (0) -+ (JR. n ) be continuous and let
[( Ql (t) )ij] be the usual matrix representation of Ql (t). Assume that
(P1 (t))ij = It"(Ql(s))ijds exists for all i, j. Let q(t) E C([to, oo),IR)
and suppose p( t) = ItCXJ q( s )ds exists with
(iv) p(fot) 2:
fo
maXij(PJ(t))ij, t E [to, (0).
If equation (7.4.21) has a nonoscillatory solution on [to, (0), then so does

(7.4.2).
Proof. Let x(t) be a nonoscillatory solution of equation (7.4.21) and let
u(t) = x'(t)x- 1 (t) to obtain the Riccati equation u'(t) = -q(t) - u 2(t).
Oscilla,tion results for differential systems
547
Put vet) = (1/ y'Ti)u( y'Tit), to get
V'(t) = - q( y'Tit) - nv 2 (t).
(7.4.22)
Integrate (7.4.22), to find
v(t) = v(T)
IT
q(y'Tis)ds + n
IT
v 2 (s)ds.
(7.4.23)
Letting T --+ 00 in (7.4.23), it follows that limT->oo veT) = b,

- 00 s; b < 00. It is easy to see that 0 s; b < 00, and we have
vet)
1
b + y'Tip(
vnt ) + n 1 v (s)ds.
t
00
where
(7.4.24)
Let Zo be the set of continuous functions Z(t) = [(Z(t))ij] from [to, (0)
to (JRn ) such that !(Z(t))ij! s; v(t), t E [to, (0) for each i, j. Define
the map Fo: Zo --+ Zo by
(7.4.25)
for t E [to, (0), i, j
= 1,2,n.
It follows from the hypothesis (iv) and (7.4.24) that Fo is well defined, Zo is closed, convex subset of M, the Frechet space of continuous
operator-valued functions on [to, (0) with the compact-open topology.
Arguments similar to those used in the proof of Theorem 7.4.2 show that
Fo is continuous on Zo and that Fo(Zo) is precompact subset of Zoo
Hence Tychonov's theorem yields the existence of a fixed point of F o, i.e.,
there exists Z(t) = [(Z(t))ij] such that
(Z(t))ij = (P1(t))ij
for t E [to, (0), i, j
1 (~(Z(S))ik(Z(S))kj)
00
ds
(7.4.26)
= 1,2,,n.
From the matrix Riccati equation of (7.4.2), it is clear that (7.4.26)

enables us to define a nonoscillatory solution of the system (7.4.2).
Example 7.4.2.
n = 2. Let
As an application of Theorem 7.4.2 consider the case
Q(t) = t21
[ac d
b1
'
where a, b, c, d are nonnegative constants and t E [1, (0). Invoking

condition (iii), there is a nonoscillatory solution X(t) of (7.4.1) satisfying
548
Chapter 7
(ii) if and only if the sequence {Pn(t)} converges as n ~
Po(t) = P(t) = -1
t
[ac
b
d
00,
where
1,
If
we obtain the relations
(7.4.27)
Since a, b, e, d::::: 0, the sequences an, bn , Cn, dn are increasing with

n, and hence have finite limits Q, (3, " 6 if and only if the system of
equations
(7.4.28)
has a solution.
It can be shown that the solvability of (7.4.28) is equivalent to the
existence of positive solution m of the equation
(7.4.29)
and
Q,
(3, ,
and 6 are given by

Q
[l_m_ d: a], (3 !
=~, 6= ~ [l-m+ d:a].
~
This in turn is easily seen to be equivalent to the condition

Q
{
< 1/4
if d
+ a + J(d -
=a
a)2
and
+ 4be
be
=0
:::: 1/2
if d
f. a
or
bc
f. O.
(7.4.30)
549
Let a, b, c, d be arbitrary nonnegative constants satisfying (7.4.30)

with equa1ity. Further, let Ql(t) = [%(t)] be chosen so that
lilllsup t1
{
t-'>oo
00
qll(s)ds < a,
limsuptl q21(S)ds <

1.-'>00
C,
lim sup t 1
t-'>oo
lim sup t 1
t-'>oo
00
q12 (s )ds
00
q22( s )ds < d.
<b
(7.4.31 )
Then by Theorem 7.4.2 the system (7.4.2) has a llonoscillatory solution.

This example may be thought of as an extension to the 2 x 2 matrix
case of Hille's Ilonoscillation criterion
1
limsup tql(t) < t--> CXJ
(7.4.32)
for the scalar equation xl/(t) + ql(t)X(t) = O. In fact, in the case b = c

d = 0, we have a = 1/4 and (7.4.31) reduces to (7.4.32).
7.5. Oscillation Results for Differential

Systems with Forcing Terms
Consider the system
XI/(t)
+ Q(t, X(t)) = F(t).
(7.5.1)
Let Mn denote the space of all n x n matrices and assume that

(i)
Q: ltD, (0) x Mn --+ Mn is continuous and
IIQ(t, X)II ::; q(t,IIXII),
(t, X) E lt~, (0) x M
n ,
to 20
where q E C([to,oo) X JR+,JR+) and q(t,z) is increasing in z,

(ii) F: [to, (0) --+ Mn is continuous and there exists a continuous V: [to,
(0) --+ Mn such that F(t) = VI/(t), t E [to, (0)The following theorem guarantees the existence of solutions of (7.5.1)
with a certain asymptotic behavior.
Theorem 7.5.1. Let conditions (i) and (ii) hold, and assume that for
some constant ,\ > 0,
CJ
sq(s,,\ + IIV(s)ll)ds <
00.
(7.5.2)
to
Then if K E Mn with IIKII <,\, there exists a solution X(t) of the

system (7.5.1) which satisfies limHOO[X(t) - V(t)] = K.
550
Chapter 7
Proof. Let X(t) be a solution of (7.5.1) and set Y(t) = X(t) - V(t).
Then it follows from the system (7.5.1) that
yl/(t)
+ Q(t, Y(t) + V(t))
o.
(7.5.3)
We shall show that the integral equation
Y(t) = K
.[XJ (t- s)(2(s, Y(s) + V(s))ds
(7.5.4)
has a solution on [t l , (0) for some tl > to. For this, it suffices show that
the operator
(TY)(t) = K
00
(t - s)Q(s, Y(s)
+ V(s))ds,
has a fixed point in a suitable Banach space of

Let t E [to, (0) and consider
BE
{u: [t, (0)
--t
~ tl
matrix~valued
(7.5.5)
functions.
Mn; U is continuous and
lim U (t ) exis ts as a finite matrix} .
t-....}CX)
Clearly, B{ is a Banach space under the sup norm. From (7.5.2), there
exists a h > to such that for every t ~ tl,
IIKII + 100J (s -
t)q(s, A + IjV(s)ll)ds ::-:: A.
(7.5.6)
Denoting S = {U E Bt,: 11U11 : -: A}, we get from (7.5.6) that T(S) c S.

Now, it is easy to see that the set T(S) is equicontinuous. In fact, this
is a consequence of the fact that every function in T(S) has derivative
bounded above, for all t ~ t 1 , by the integral
00
q(s, A + IjV(s)ll)ds <
It is also easy to see that T is continuous and if YES,
IITY - KII <
00
(7.5.7)
00.
then
(s - t)q(s, A + IjV(s)ll)ds.
(7.5.8)
Since the integral on the right~hand side of (7.5.8) tends to zero independently of Y, it follows that the set T( S) is equiconvergent. Consequently,
T(S) is relatively compact in Bt,. Thus, by Tychonov's theorem, the operator T has a fixed point in the ball S. This fixed point Y (t) satisfies
equation (7.5.1). The conclusion of the theorem now follows by letting
X(t) = Y(t) + V(t).
551
Next, we employ Theorem 7.5.1 to obtain an infinity of oscillatory solutions of the system (7.5.1).
Theorem 7.5.2. Let conditions (i) and (ii) hold, and assume that for
some A > 0 condition (7.5.2) is satisfied, and
(iii) IW(t)11 is bounded on [t1' (0), t1 > to ::::> 0,

(iv) there exist L E Mn and 5> 0 such that IILII
limsnp det[L
t->oo
and
liminf det[L
t->oo
+ V(t)]
<)..
and
> 5,
+ V(t)] < -
(7.5.9)
5.
(7.5.10)
Then there exists an oscillatory solution of the system (7.5.1).

Proof. Theorem 7.5.1 implies the existence of a solution X(t) of (7.5.1)
such that X(t) - V(t) -+ L as t -+ 00. We can assume that this solution
X(t) is defined for all t > t1' Let {tj},] = 1,2""
be such that
limj->oo tj = 00, and det[L + V(tj)] = 5, ] = 1,2" ". Since V(t) is
bounded, X(t) is also bounded on [t1'(0) and this implies the existence
of a compact set Z in Mn such that V(t), X(t) E Z, t E [t1' (0).
By the uniform continuity of the determinant function on Z, there exist
p( 5) > 0 and a positive integer ]0 such that
(7.5.11)
and
(7.5.12)
for every] ::::> ]0. Consequently, from (7.5.12) we deduce that det(X(tj)) >
5/2 for all ]::::> ]0. Similarly, we can show that det(X(si)) < -5/2 for
is a sequence in [t1' (0) with
all large i, where {Si}, i = 1,2""
limi->oo Si = 00. Obviously, det(X(t)) has an unbounded set of zeros on
[t1' (0) and this completes the proof.
Example 7.5.1.
In the system (7.5.1) let
Q(t, X)
(t
+ 1)311XII
and
F(t) =
sint
1/(t + 1)
1/(t+l) ]"
cost
It is easy to see that the hypotheses of Theorem 7.5.2 are satisfied, and
hence (7.5.1) has at least one oscillatory solution.
Chapter 7
552
Remark 7.5.1. It is worth noting that if the hypotheses of Theorem 7.5.2

are satisfied, there is an infinity of oscillatory solutions of (7.5.1). This
follows easily from the fact that the integral equation (7.5.4) has solutions
for all 'small' matrices K.

1. Theorems 7.1.1,7.1.3 and 7.1.4 are taken from Kwong and Wong [33].
These theorems extend the sufficiency part of the results due to Atkinson
[5], Belohorec [7] and Waltman [49). Lemma 7.1.1 and Theorem 7.1.2 are
due to Kordonis and Philos [27]. For some other related work we refer to
Mirzov [39-41].
2. The results presented in Section 7.1 can be extended to systems of
type (7.1.1) (and its prototype the system (7.1.2)) with deviating arguments, namely,
x' (t) = q1 (t)fI (Y[91 (t)])

y'(t) = - q2(t)!2(X[92(t)]) ,
where qi and 1;, i = 1,2 are as in (7.1.1) with qi(t) 2: 0 for t 2:
to, i = 1,2 and 9i(t) E C([to, 00), IR), limHoo 9i(t) = 00, i = 1,2. The
formulation of these results are left to the reader.
3. Another possible extension is the study of the system (7.1.2) when

A1A2 = 1. This is so called half-linear case. With the introduction of
the variable w (t) = - (Y2 (t) / X>'2 (t) )sgn x (t), we get the Riccati equation w'(t) = q2(t) + A2q1(t)W>'(t), where A = (A2 + 1)/A2. All Sturmian
comparison-type theorems continue to hold and oscillation criteria can easily be obtained via the traditional methods (see Chapter 3).
4. Theorem 7.2.1 is due to Mingarelli [37,38], whereas Theorem 7.2.2

and Corollaries 7.2.1 - 7.2.4 are from Kwong et. a1. [29]. Criteria (id (i4 ) are extracted from the work of Kwong and Zettl [34,35]. Lemmas 7.2.1,
7.2.2 and Theorems 7.2.3 and 7.2.4 are taken from Butler and Erbe [10].
Lemmas 7.2.3, 7.2.4 and Theorems 7.2.5 - 7.2.10 are borrowed from Butler
et. a1. [11].
5. The scalar Riccati inequality (7.2.19) remains valid if instead of the
trace, we use any positive continuous linear functional on the space of real
symmetric n x n matrices to define the function w(t). Applying such a
functional , say, to both sides of the equation (7.2.10), we obtain, instead
of (7.2.11),
(W(t))
(W(to)) + (Q1(t)) +
it
(W2(s))ds.
(7.6.1)
to
It has been shown by Akiyama [3], that any continuous positive linear
functional on the cone of nonnegative real symmetric matrices is equivalent
553
with the trace functional. Hence, there exists a positive constant c, which
depends only on , such that (W2(t));c c tr(W 2(t)). Thus, instead of
(7.2.13) we have the inequality
(W2(t)) ;c c(l-tlAi(Ql(t))
forall
t;cT.
(7.6.2)
The analogue of (7.2.14) is
t (W2(s))ds
lto
= f
t (W2(s))ds + (1 - f) ltot (W2(s))ds.
lto
(7.6.3)
Here we use (7.6.2) to establish the last integral, whereas in the first integral
we use the elementary inequality (W2(t));c ((J))-l[(W(t))j2, where
J is the n x n identity matrix. Thus, we obtain instea.d of (7.2.17),
(W(t))
> (Ql(t)) + (W(to)) + c(l- f)3
+ ~I)
1:
Ai(Ql(S))ds
(7.6.4)
[(W(s)Wds.
;c c(l for t;c Tl ;c T. Thus if we let 8 = c(l define w(t) = (W(t)), t;c to then
q(t) = (Ql(t))
+8
lto
f)4,
f)4
t Ai(Ql(S))ds (7.6.5)
lto
replace
by (I)E, and
Ai(Ql(S))ds.
Hence, inequality (7.2.19) holds again.

Therefore, at least in some results, we can replace the functional trace
by any positive linear functional in the space of real symmetric n x n
matrices. For a more general discussion of this we refer the reader to
Akiyama [3] and Etgen and Pawlowski [18,19].
It:
6. Theorem 7.2.3 can be generalized by considering the principal subQ(s)ds. We recall (cf. 6], p. 113): For any n x n symmatrices of
metric matrix A, the sequence of symmetric matrices Ak = (aij), i, j =
1,2, ... ,k for k = 1,2, ... ,n satisfies A]+l (Ak+1) ::; Aj (Ak) ::; Aj (A k + 1),
where Aj(Ak) denotes the j th characteristic roots of A k .
Theorem 7.2.3'. Let 9 = g(t) be positive, continuous and nondecreasing

in [to, (0) and assume there exists k, 1 ::; k ::; n such that
limapproxinf (1) tr
t -+ 00
9 t
(itto QdS)dS)
> -
00,
(7.6.6; k)
Chapter 7
554
00,
and
lim)'1
t--+oo
(fiQds)ds)
. to
00.
(7.6.7; k)
(7.6.8; k)

Proof.
have
The proof proceeds as in Theorem 7.2.3 upto (7.2.33). Then, we
and therefore Al (- Vk (t)) -+ 00 as t -+ 00 by (7.6.8;k). A straight-forward

modification of the proof of Theorem 7.2.3 now yields a contradiction to
condition (7.6.7; k).
7. Lemma 7.3.1 is the Theorem 8 in [20], whereas Lemmas 7.3.2 and

7.3.3 can be found in [16]. The rest of the results in Section 7.3 are taken
from Butler and Erbe [9], except Corollary 7.3.2 and the two criteria (Id
and (12) which are due to Kura [28]' Macki and Wong [36]' and Belohorec
[7], respectively. We also remark that Theorems 7.3.6, 7.3.7 and 7.3.8 are
the extensions of the results of Macki and Wong [36], Belohorec [7], and
Atkinson [5], respectively to second order matrix equations.
8. The results of Section 7.4 are taken from Butler and Erbe [8]. We
note that Theorem 7.4.1 compares two operator-valued equations, whereas
the results of Etgen and Lewis [17] compare an operator-valued equation
with a scalar equation. We also refer the reader to the papers [1,2] for some
comparison results of nonself-adjoint equations.
9. The results of Section 7.5 are taken from Kartsatos and Walters [26].
7.7. References
1. S. Ahmad and A.C. Lazer, A n-dimensional extension of the Sturm
separation and comparison theory to a class of nonsclf-adjoint systems,
SIAM J. Math. Anal. 9(1978), 1137-1150.
2. S. Ahmad and A.C. Lazer, On an extension of Sturm's comparison theorem to a class of nonself-adjoint second order systems, Nonlinear Analysis
4(1980), 497-501.
3. K. Akiyama, On the maximum eigenvalue conjecture for the oscillation
of second order differential systems, M.Sc. Thesis, University of Ottawa,
1983.
555
4. W. Allegretto and L. Erbe, Oscillation criteria for matrix differential

inequalities, Canad. Math. Bull. 16(1973),5-10.
5. F.V. Atkinson, On second order nonlinear oscillation, Pacinc J. Math.
5(1955), 643-647.
6. R. Bellman, Introduction to Matrix Analysis, 2nd ed. McGraw Hill, New
York, 1970.
7. S. Belohorec, Oscillatory solut.ions of certain nonlinear differential equations of second order, I\i[at. Fyz. Casopis Salven. Akad. Vied. 11(1961),
250-255.
8. G.J. Butler and L.H. Erbe, Comparison theorems for second order
operator-valued linear differential equations, Pacinc J. Math. 112(1984),
21-34.
9. G.J. Butler and L.H. Erbe, Oscillation theory for second order differential systems with functionally commutative matrix coefficients, Funkcial.
Ekvac. 28(1985), 47-55.
10. G.J. Butler and L.H. Erbe, Oscillation results for second order differential systems, SIAM J. Math. Anal. 17( 1986), 19-29.
11. G.J. Butler, L.H. Erbe and A.B. Mingarelli, Riccati techniques and
variational principles in oscillation theory for linear systems, Trans. Arner.
Math. Soc. 303(1987), 263-282.
12. R. Byers, J. Harris and M.K. Kwong, Weighted means and oscillation
conditions for second order matrix differential equations, J. DiHerential
Equations 61(1986), 164-177.
13. W.J. Coles, Oscillation criteria for nonlinear second order equations, Ann.
Mat. Pura. Appl. 82(1969), 123-134.
14. W.J. Coles and D. Willet, Summability criteria for oscillation of second
order linear differential equations, Ann. Mat. Pura. Appl. 79(1968),391398.
15. W.A. Coppel, Disconjugacy, Lecture Notes in Math. 220, SpringerVerlag, New York, 1971.
16. M.P. Drazin, J.W. Dungey and K.W. Greuenberg, Some theorems
on commutative matrices, J. London ]V]ath. Soc. 25(1950), 221-228.
17. G.J. Etgen and R.T. Lewis, A Hille-Wintner comparison theorem for
second order differential systems, Czech. Math. J. 30(1980), 98-107.
18. G.J. Etgen and J.F. Pawlowski, Oscillation criteria for second order
selfadjoint differential systems, Pacinc J. Math. 66(1976),99-110.
19. G.J. Etgen and J.F. Pawlowski, A comparison theorem and oscillation
criteria for second order differential systems, Pacinc J. Math. 72(1977),
59-69.
20. H.I. Freedman, Functionally commutative matrices and matrices with
constant eigenvectors, Linear and Multilinear Algebra 4(1976), 197-213.
1964.
22. T.L. Hayden and H.C. Howard, Oscillation of differential equations in
Banach spaces, Ann. Mat. Pura Appl. 85(1970), 383-394.
556
Chapter 7
23. E. Hille, Nonoscillation theorems, Trans. Arner. Math. Soc. 64(1948),

234-252.
24. E. Hille, Lectures on Ordinary Differential Equations, Addison-Wesley,
Reading, Mass. 1969.
25. D. Hinton and R.T. Lewis, Oscillation theory for generalized second
order differential equations, Rocky Mountain J. Math. 10(1980),751-766.
26. A.G. Kartsatos and T. Walters, Some oscillation results for matrix
and vector differential equations with forcing term, J . .Math. Anal. Appl.
73(1980), 506-513.
27. L-G.E. Kordonis and Ch.G. Philos, On the oscillation of nonlinear
two-dimensional differential systems, Proc. Arner. Math. Soc. 126(1998),
1661-1667.
28. T. Kura, A matrix analog of Atkinson's oscillation theorem, Funkcial.
Ekvac. 28(1985), 47-55.
29. M.K. Kwong, H. Kaper, K. Akiyama and A. Mingarelli, Oscillation of linear second order differential systems, Proc. Arner. IVIath. Suc.
91(1984), 85-91.
to ::;econd order nonlinear oscillation, J. DiHerential Equations 46(1982),
63-77.
31. M.K. Kwong and J .S.W. Wong, Linearization of ::;econd order nonlinear oscillation theorems, Trans. Arner. Math. Soc. 279(1983), 705-722.
32. M.K. Kwong and J.S.W. Wong, On the oscillation theorem of Belohorec, SIAM J. Math. Anal. 14(1983), 474-476.
33. M.K. Kwong and J.S.W.Wong, oscillation of Emden--Fowler systems,
DiHerential and Integral Equations 1(1988), 133-141.
34. M.K. Kwong and A. Zettl, Integral inequalities and second order linear
oscillation, J. DiHerential Equations 45(1982), 16-33.
35. M.K. Kwong and A. Zettl, Asymptotically constant functions and second order linear oscillation, J. Math. Anal. Appl. 93(1983), 475-491.
36. J .W. Macki and J .S.W. Wong, Oscillation of solutions to second order
nonlinear differential eqnations, Pacific J. Math. 24( 1968), Ill-ll7.
37. A.B. Mingarelli, An oscillation criterion for second order self adjoint
differential systems, G.R.. Math. Rep. Acad. Sci. Canada 2(1980), 287290.
38. A.B. Mingarelli, On a conjecture for oscillation of second order ordinary
systems, Proc. Arner. Math. Soc. 82(1981), 593-598.
39. D.D. Mirzov, On oscillatoriness of the solutions of a system of nonlinear
differential equations, DiHerencial'nye Uravnenija 9(1973), 581-583.
40. D.D. Mirzov, The oscillation of solutions of a system of nonlinear differential equations, Math. Zarnetki 16(1974), 571-576.
41. D.D. Mirzov, Oscillation properties of solutions of a nonlinear EmdenFow ler differential system, DiHerencial'nye Ura vnenija 16( 1980), 1980-1984.
42. M. Morse, The Calculus of Variations in the Large, Arner. I'vIath. Soc.
Colloq. Publ. 18, New York, 1934.
557
43. C. Olech, Z. Opial and T. Wazewski, Sur Ie probleme d'oscillation des

integrales de l'equation y" + q(t)y = 0, Bull. Akad. Polan. Sci. 5(1957),
621-626.
44. B.N. Parlett, The Symmetric Eigenvalue Problem, Prentice-Hall, Englewood Cliffs, NJ, 1980.
45. A.P. Robertson and W. Robertson, Topological Vector Spaces, 2nd
ed. Cambridge University Press, Cambridge, 1973.
46. H.H. Schaefer, Banach Lattices and Positive Operators, Springer-Verlag,
New York, 1974.
47. E.C. Tomastik, Oscillation of systems of second order differential equations, J. Differential Equations 9(1971), 436-442.
48. T. Walters, A characterization of positive linear functionals and oscillation criteria for matrix differential equations, Proc. Amer. Math. Soc.
78(1980), 198-202.
49. P. Waltman, An oscillation criterion for a nonlinear second order equation, J. Math. Anal. Appl. 10(1965), 439-441.
50. D. Willet, Classification of second order linear differential equations with
respect to oscillation, Advances Math. 3(1969), 594-693.
51. D. Willet, A necessary and sufficient condition for the oscillation of some
linear second order differential equations, Rocky Mountain J. Math. 1
(1971), 357-365.
52. A. Wintner, On the comparison theorem of Kneser-Hille, Math. Scand.
5(1957), 255-260.
53. J.S.W. Wong, On two theorems of Waltman, SIAM J. Appl. Math.
14(1966), 724-728.
equations, Proc. Amer. Math. Soc. 106(1989), 1069--1077.
55. J.S.W. Wong, Oscillation criteria for second order nonlinear differential
equations with integrable coefficients, Proc. Amer. Math. Soc. 115(1992),
389--395.
Chapter 8
Asymptotic Behavior
of Solutions of Certain
8.0. Introduction
The study of behavioral properties of solutions of differential equations
near infinity is of immense importance and hence it continues to attract
many researchers. Therefore, in this chapter we shall present some recent
contributions on the asymptotic behavior of solutions of second order differential equations as well as the behavioral properties of positive solutions
of singular Emden-Fowler-type equations. In Section 8.1 it is shown that
for a large class of differentia'! equations, not only can the existence of
nonoscillatory solutions be proved, but also an explicit asymptotic form of
the nonoscillatory solutions may be provided. Then, we shall impose more
restrictions on the sign of the integrable coefficient of the equation, and get
necessary and sufficient conditions so that the solutions have the specified
asymptotic behavior as t -+ 00, i.e., solutions which behave asymptotically like a nonzero constant and a.!so those which behave asymptotically
like et, e i- O. For this, various averaging techniques of the types employed in the previous chapters to study the oscillatory behavior of such
equations have been used. Section 8.2 is devoted to the study of existence, uniqueness and asymptotic behavior of positive solutions of singular
Emden-Fowler-type equations. The cases when the coefficient of the equation under consideration is of constant sign, or of an alternating sign are
systematically discussed. Then the existence as well as nonexistence results
for the positive solutions of Emden-Fowler-type systems are proved.
8.1. Asymptotic Behavior of Solutions of

Here we shall study the asymptotic behavior of solutions of the second
Asymptotic behavior of solutions of certain differential equations
559
order nonlinear differential equation
x"(t)
+ q(t)f(x(t)) =
(8.1.1)
0,
where q(t) E C([to,oo),lR), f(x) E C1(lR,lR), to > 0 and f satisfies

xf(x) > 0 and f'(x):2: 0 for all xi- O. When f(x) = Ixll'sgn x, x E
lR b > 0), equation (8.1.1) reduces to
x"(t) + q(t)lx(t)ll'sgn x(t) = O.
(8.1.2)
The differential equation (8.1.2) is the prototype of (8.1.1) and is known as

the generalized Emden-Fowler equation.
8.1.1. Asymptotic Behavior of Nonoscillatory Solutions

In the study of the asymptotic behavior of solutions of differential equations of the form (8.1.1) an interesting problem is to establish necessary
and/or sufficient conditions for the existence of solutions which behave like
nontrivial linear functions Cl + c2t as t --+ 00. Such a solution satisfies
the asymptotic condition
x(t) = c+o(l)
x(t) = o(t)
or, the condition
and
as
lim x(t) =
x(t) = ct + o(t)
as
t--+oo,
00
as
t --+
(8.1.3)
t --+
00,
00,
(8.1.4)
(8.1.5)
where c is a nonzero constant.

In this subsection, we shall discuss the asymptotic properties of nonoscillatory solutions of (8.1.2) when
lim
t-400
it
to
q(s)ds exists and is finite.
(8.1.6)
If condition (8.1.6) holds, then we can define the function Q by
Q(t)
00
q(s)ds,
t:2: to.
(8.1.7)
If q( t) :2: 0 for t :2: to, then it is easy to show that a nonoscillatory solution
x(t) of (8.1.1) satisfies exactly one of the three asymptotic conditions (8.1.3)
- (8.1.5). In the following result, we shall prove that this fact remains valid
even when Q(t):2: 0 for t:2: to
Theorem 8.1.1.
Let
r > O. Suppose in addition to condition (8.1.6),
Q(t) :2: 0 for t:2: to. Then for each nonoscillatory solution x(t) of
Chapter 8
560
equation (8.1.2), exactly one of the three asymptotic conditions (8.1.3) (8.1.5) is satisfied.
x(t) > 0 for t 2: to > O. It is known (cf. Lemma 4.1.3) that x(t) satisfies
Proof.
the equality
x'(t)
= Q(t)x'Y(t) +ryx'Y(t)
00
x-'Y-1(s)(x'(s)fds
for
2: to,
where Q(t) is defined in (8.1.7). Therefore, we have
x'(t) 2: Q(t)x'Y(t)
t 2: to.
for
(8.1.8)
Since Q(t) 2: 0 for t 2: to, it follows that x'(t) 2: 0 for t 2: to. An

integration by parts of equation (8.1.2) gives
x'(u)-Q(u)x'Y(u)+ry
Q(s)x'Y-l(s)x'(s)ds
x'(t)-Q(t)x'Y(t), (8.1.9)
where u 2: t 2: to. Let t be fixed. Since Q(s)x'Y-1(s)x'(s) is nonnegative,

the integral term in (8.1.9) has a finite limit, or diverges to 00 as t --+ 00.
If the latter case occurs, then x'(u) - Q(u)x'Y(u) --+ -00 as u --+ 00,
which is a contradiction to (8.1.8). Thus, the former case occurs, i.e.,
/00 Q(s)x'Y-l(s)x'(s)ds
<
00.
(8.1.10)
Define the function kl (t) as

(8.1.11)
and the finite constant
(8.1.9) yields
x'(t)
0:
0: =
limu .... oo[x'(u) - Q(u)x'Y(u)]. Then equality
+ Q(t)x'Y(t) + ryk1(t)
Observe by (8.1.8) that
0:
for
2: to.
(8.1.12)
2: O. From (8.1.8) and (8.1.10) it follows that
/00 Q2(s)x2'Y-l(S)ds
<
00.
(8.1.13)
Next, we define the function k2(t) by

(8.1.14)
561
Integrating the equation (8.1.12) from to to t, we get
x(t) = x(to)
+ a(t -
to)
t Q(s)'i(s)ds + "( itot kl(S)ds.
ito
By Schwartz's inequality and the fact that x'(t);:::: 0 for

integral term in (8.1.15) can be estimated as follows
t Q(s)x'(s)ds <
ito
(1:
Q2(S)x 2,-1(s)dS) 1/2
< [k2(tO)(t - to)X(t)J1/2
X(S)dS) 1/2
the first
(8.1.16)
t;:::: to.
for
t;:::: to, we find
Thus, for
x(t)
(1:
t;:::: to,
(8.1.15)
:s
x(to)
+ a(t -
+ [k2(tO)(t -
to)
to)X(t)J1/2
+ "(kI (to)(t -
The above inequality may be regarded as a quadratic inequality in

Then, we have
X1/ 2(t)
:s ~
[(k 2(t)(t - to))1/2
+ D1/2(t)]
for
to).
X 1/ 2
(t).
t;:::: to,
where D(t) = k2(tO)(t - to) + 4[x(to) + a(t - to) + "(k(to)(t - to)]. It is

obvious that D(t) = O(t) as t ---+ 00, and consequently, there exists a
positive constant m such that
x(t) :S mt
for
t;:::: to.
(8.1.17)
Let T;:::: to be an arbitrary number. It is clear that
o :S -lit Q(s)x'(s)ds
t
liT Q(s)x'(s)ds + -lit Q(s)x'(s)ds
J;
for t;:::: T. Arguing as in (8.1.16), we find

Q(s)x'(s)ds :S [k2(T)(tT)x(tW/ 2 for t;:::: T, which when combined with (8.1.17) yields
Q(s)x'(s)ds :S [mk2(T)t(t - T)P/2
Taking the upper limit as t ---+

we obtain
o :S
to
[mk2(T)p/2.
Since T is arbitrary and k2(T) ---+ 0 as T ---+

(8.1.19), we get
lit Q(s)x'(s)ds
lim -
t--400
to
t;:::: T.
(8.1.18)
in the above equality and using (8.1.18),
lit Q(s)x'(s)ds:S
lim sup t--400
00
for
00,
O.
(8.1.19)
letting T ---+
00
in
(8.1.20)
Chapter 8
562
In view of (8.1.15), (8.1.20) and the fact that kl(t)
find limt-HXJ x(t)/t = Q:.
0 as t
~ 00,
we
Recall that x(t) is nondecreasing for t:2: to. Now, there are three
cases to consider:
(i)
(ii)
(iii)
Q:
Q:
Q:
x(t) is bounded above,

0 and x(t) is unbounded,
> 0 (and hence x(t) is unbounded).
= 0 and
Case (i) implies (8.1.3) with c = limHoo x(t) > 0, while case (iii) implies
(8.1.5) with c = Q: > O. It is also clear that case (ii) implies (8.1.4). This
8.1.2. Bounded Asymptotically Linear Solutions

In this subsection, we shall present necessary and/or sufficient conditions for (8.1.1) to have solutions which behave asymptotically like nonzero
constants.
/00 IQ(s)lds
<
00
(8.1.21)
and
(8.1.22)
Then for any constant c
that
x(t)
#- 0,
equation (8.1.1) has a solution x(t) such
c+ 0
(1
00
[IQ(s)1
and
x'(t)
as t ~
00,
O(IQ(t)1
+ pes)] dS)
+ pet)),
(8.1.23)
(8.1.24)
where pet) = ftoo Q2(s)ds.
Proof. Condition (8.1.22) implies that pet) is nonincreasing and integrable on [to, 00). We may assume that c> O. Let
and choose b, T:2: to so that
mm'
+ bm'
loo
IQ(s)lds < b
(8.1.25)

and
m
Let F
{OO
iT
IQ(s)lds
(OO
+ b iT
P(s)ds ~
2".
563
(8.1.26)
be the FnChet space of continuously differentiable functions on
[T,oo) with the family of seminorms {II Ill, = 1,2,} defined by

Ilxlll = sup{lx(t)1 + Ix'(t)1 : T ~ t ~ T + }. We have the convergence
Xk -+ x (k -+ 00) in the topology of F if and only if Xk(t) -+ x(t) (k -+ 00)
and x~(t) -+ x'(t) (k -+ 00) uniformly on every compact subinterval of
[T, 00 ). Let X denote the set
X
{x
F : Ix(t) - cl
~~,
Ix'(t)1
~ mQ(t) + bP(t)
for t 2': T}.
Note that X is a nonempty closed convex subset of F. For t 2': T,

define the operator on X as follows
(x)(t) = c
-1
00
Q(s)f(x(sds
-1 (1
00
We shall find a fixed point of in

Tychonov theorem.
00
we
Q(u)f'(x(u))X'(U)dU) ds.
(8.1.27)
by employing the Schauder-
(It) is well-defined on X and maps X into itself. For this, let x E X.
For r 2': s 2': T, we have
11T Q(u)f'(x(ux'(U)dU\
<
<
lT IQ(u)lm' [mIQ(u)1 + bP(u)] du

mm' lT Q2(u)du + m'bP(s) lT IQ(u)ldu.
f;
Therefore, letting T -+ 00 and using (8.1.25), we find that

Q(u)f'(x(u
x x'(u)du converges and satisfies I f800 Q(u)f'(x(ux'(u)dul ~ bP(t) for
s 2': T. This implies
I(cpx)'(t) I <
IQ(t)f(x(t1
\100
Q(U)f'(X(Ux'(U)dU\
< mIQ(t)1 + bP(t) for t 2': T,

and in view of (8.1.26), for t 2': T, we have
1(x)(t)-cl
~ ooIQ(S)f(x(sldS+ 00
1100
Q(u)f'(x(ux'(u)dul ds
~ ~.
Thus, is well-defined on X and maps X into X.
(12 ) cp is continuous on X. For this, let x, Xk (k = 1,2,.) be functions

such that Xk(t) -+ x(t), x~(t) -+ x'(t) as k -+ 00, uniformly on
in X
564
Chapter 8
every compact subinterval of [T,oo). Then, we have
1(xk)'(t) - (x)'(t) I
-:;
IQ(t)llf(Xk(t)) - f(x(t))1
1 IQ(u)IIf'(Xk(U))X~(U)
00
- f'(x(u))x'(u)ldu
for t 2: T. Observe that f(xk(t)) -4 f(x(k)) as t -4 00, uniformly on

compact subintervals of [T,oo) and IQ(u)llf'(xk('U,))x~(u)- f'(x(u))x'(u)1
is bounded above by J;' 2m'IQ(u)l[mIQ(u)1 + bP(u)]du and IQ(u)1
x 1f'(Xk(U))X~(u) - f'(x(u))x'(u)1 -40 as k -4 00. Applying the Lebesgue
dominated convergence theorem, we find that (Xk)'(t) -4 (x)'(t) as
k -4 00 uniformly on every compact subinterval of [T,oo). Moreover,
since
1(Xk)(t) - (X)(t) I
-:;
1I
00
(xd(s) - (x)'(s)lds
and I(xk)'(S) - (x)'(s)1 is bounded above by J;' 2[mIQ(s)1 +bP(s)]ds.

Once again by applying the Lebesgue dominated convergence theorem, we
conclude that (xk)(t) -4 (x)(t) as k -4 00 uniformly on every finite
subinterval of [T,oo). Thus, is continuous on X.
(13) X is compact. This can be proved by using Arzela-Ascoli's theorem.

Since the proof is standard (see the proofs of Theorems 4.3.1 and 4.3.4).
The details are left to the reader.
Now the Schauder-Tychonov fixed point theorem guarantees the existence of a fixed point x E X of . It can be easily verified that x = x(t)
is a solution of equation (8.1.1) for t 2: T, and satisfies the properties
(8.1.23) and (8.1.24). This completes the proof.
Remark 8.1.1. In Theorem 8.1.2 the assumption that f'(x) ~ 0 for

x -lOis unnecessary. In fact, this sign condition of f' is not used in the
proof.
It is known that if conditions (8.1.21) and (8.1.22) are satisfied, then
(8.1.1) has a nonoscillatory solution (see Theorem 4.3.1). Thus, Theorem
8.1.2 asserts the existence of a nonoscillatory solution of (8.1.1) with the
asymptotic behavior (8.1.23) and (8.1.24).
x"(t)
+ (ktA sin t)lx(t)I'"Y
sgn x(t)
0,
(8.1.28)
where k, A and 'Y > 0 are constants. Applying Theorem 8.1.2 to the
case f(x) = Ixl'"Y sgn x, IQ(t)1 = too ks Asinsdsl -:; 21klt A (). < 0), we
find that if A < -1, then for any nonzero constant c, equation (8.1.28)
has a solution x(t) such that x(t) = c + O(tA+l) and x'(t) = O(tA) as
t -4 00. Note that (8.1.28) has a nonoscillatory solution if and only if
IJ

(i)
565
>. < -1 for, > 1 (see [1-3]),
(1'1') { >. < -1, k arbitrary

>. = -1, Ikl :s: 1/J2
f
or,
1 (
[415])
see,
,
>. < -, for 0 < , < 1 (see [4,5]).

The superlinear case h > 1) shows that Theorem 8.1.2 is 'sharp' in
the sense that (8.1.28) never has a nonoscillatory solution unless >. < -1.
(iii)
The following corollaries of Theorem 8.1.2 are immediate.

Corollary 8.1.1. Suppose conditions (8.1.21) and (8.1.22) are satisfied.
Then equation (8.1.1) has a nonoscillatory solution x(t) satisfying (8.1.3).
Corollary 8.1.2. Suppose condition (8.1.21) and
lim tQ(t)
t--too
(8.1.29)
are satisfied. Then equation (8.1.1) has a nonoscillatory solution x(t) such
that for any constant c =I 0,
x(t)
c+o(l)
and
x'(t)
o(C I )
as
t---+oo.
(8.1.30)
Proof.
It suffices to note that conditions (8.1.21) and (8.1.29) imply
OO
(8.1.22) and t OO Q2(s)ds
sQ2(s)ds ---+ 0 as t ---+ 00.
ft
:s: ft
In what follows it will be shown that the converse of Corollaries 8.1.1

and 8.1.2 in some sense is possible provided Q(t) is of fixed sign.
Theorem 8.1.3. Suppose Q(t) ~ 0 for all large t and f'(x) > 0 for
x =I O. Then the following statements are equivalent:
(i)
for any constant c =I 0, there exists a solution x(t) of equation
(8.1.1) satisfying (8.1.3),
(ii) for some constant c =I 0, there exists a solution x(t) of equation
(8.1.1) satisfying (8.1.3),
(iii) the integral conditions (8.1.21) and (8.1.22) are satisfied.
Proof. (i) implies (ii) trivially, and (iii) implies (i) by Corollary 8.1.1. We
claim that (ii) implies (iii). Let x(t) be a solution of equation (8.1.1) for
which (8.1.3) holds for some constant c =I O. We may assume that c> O.
There is a number T such that c/2:S: x(t) :s: 2c for t ~ T. Recall
Lemma 4.1.3, if we remove condition (4.1.49) and let a(t) = 1, then the
equality (4.1.52) takes the form
x'(t)
f(x(t))
= a + Q(t) +
00
x'(s)
f(x(s))
)2 f ,(x(s))ds,
(8.1.31 )
Chapter 8
566
where a is a nonnegative constant. Since Q(t):::: 0,
x'(t)
f(x(t
>
- Q(t)
00
m1Q (s)ds :::: 0
(8.1.31) gives
for
t:::: T,
where m ] = min {I' (x) : c/2 ::; x ::; 2c} > O. An integration of the above
inequality over [T, t] gives
X(t)
. x(T)
It [
- du > . T
f(u) -
Q(s)
00
m]
Q 2 (u)du J ds
for
t:::: T.
Since the left-hand side of t.he above inequality remaius bouuded as t--+
00,
we conclude that (8.1.21) and (8.1.22) are satisfied. This completes
the proof.
For the superlinear case, i.e., equation (8.1.2) with "f > 1,
result is now clear.
Theorem 8.1.4.
Q(t) :::: 0 for
t::::
the next
Let "f > 1. Suppose condition (8.1.6) holds and that

to. Then the following three statements are equivalent:
(i)
equation (8.1.2) has a nonoscillatory solution x(t) satisfying (8.1.3),
(ii)
equation (8.1.2) has a nonoscillatory solution,

In fact, the equivalence of (ii) and (iii) can be obtained from the contrapositive form of a special case of Corollary 4.3.1, and the equivalence of
(i) and (iii) is given in Theorem 8.1.3.
Theorem 8.1.5. Suppose either Q(t)::::

or Q(t)::;
Then the following three statements are equivalent:
for all large t.
(i)
for any constant c
(8.1.1) satisfying (8.1.30),
0,
there exists a solution x( t) of equation
(ii) for some constant c

(8.1.1) satisfying (8.1.30),
0,
there exists a solution x(t) of equation

Proof. As (i) implies (ii) trivially, and (iii) implies (i) by Corollary 8.1.2,
it suffices to show that (ii) implies (iii). Suppose x(t) is a solution of
(8.1.1) such that (8.1.30) holds for some constant c f 0. As in the proof of
Theorem 8.1.3, we employ Lemma 4.1.3 to obtain (8.1.30) which is satisfied
for all large t. Since a = limt--+=x'(t)/f(x(t = 0, it follows that
x' (t)
f(x(t
= Q(t) +
00
x' (s ) )
f(x(s
f (x(sds,

or equivalently,
tx'(t)
tQ(t) = f(x(t - t
00
x'es)
f(x(s
)2 f ,(x(sds
567
(8.1.32)
for all large t. The term tx'(t)/ f(x(t -+ 0 as t -+ 00 since x(~)

satisfies (8.1.30). Using l'Hospital's rule and (8.1.30), we find
.1
t~! t
00
x' (s ) )
f(x(s
. ft ft C(:~:~)
OO
f (x(sds
hm
t-too
f'(x(sds
2
----'-----'-----':-'-~----
.!l..
dt
(1)
t
f'(x(t [ '( )]2
t~c!, j2(x(t tx t
O.
Therefore, we have (8.1.29).

Next, we shall show that (8.1.21) holds. In view of (8.1.1), we have
(x(t) - tx'(t' = tq(t)f(x(t

= -(tQ(t)f(x(t))' + Q(t)f(x(t
+ tQ(t)f'(x(tx'(t).
An integration of the above equality from T to t gives
x(t) - tx'(t)
0'1 -
tQ(t)f(x(t
Q(s)[f(x(s + sf'(x(sx'(s)]ds,
it
(8.1.33)
where O'} = x(T)-Tx'(T)+TQ(T)f(x(T. Since tQ(t) -+ 0 as t -+ 00

and x(t) satisfies (8.1.30), we have x(t) - tx'(t) -+ c, tQ(t)f(x(t -+ 0
and f(x(t + tx'(t)f'(x(t -+ fCc) of- 0 as t -+ 00. These facts together
with (8.1.33) and the sign property of Q(t) easily lead to condition (8.1.21).
Remark 8.1.2. From Theorems 8.1.3 and 8.1.5, we note that even for
solutions which have the same limit as t -+ 00, there is an essential
difference between restricting and not restricting the asymptotic behavior
of their derivatives.
8.1.3. Unbounded Asymptotically Linear Solutions

In this subsection we shall present necessary and/or sufficient conditions
for equation (8.1.1) to have solutions which behave asymptotically like ct (c
is a nonzero constant). In the above results, no growth condition on f
was required in proving the existence of a solution asymptotic to a nonzero
constant as t -+ 00. The situation now becomes different ~nd we shall
need to impose one of the following growth conditions on f:
j'(x) is nondecreasing for x> 0 and nonincreasing for x < 0, (8.1.34)

f'ex) is nonincreasing for x> 0 and nondecreasing for x < O. (8.1.35)
Chapter 8
568
Theorem 8.1.6. Suppose either condition (8.1.34) or (8.1.35) is satisfied.

Also, suppose for all constants k 1= 0, kl 1= 0 and k2 1= 0
t1
jt If(ks)IIQ(s)lds -+ 0 as t -+
jt f'(ks)IQ(s)lds <
and
IX) If(klS)If'(k2S)Q2(s)ds
00,
(8.1.37)
00,
<
(8.1.36)
(8.1.38)
00.
Then for any nonzero constant c equation (8.1.1) has a solution x(t)
such that
x(t)
and
as t -+
ct + 0 (jt [If(c, s)IIQ(s)1
+ Pc(S)]dS)
x'(t) = c + 0 (If(c, t)IIQ(t)1

00,
+ Pc(t)) ,
(8.1.39)
(8.1.40)
where
_ 3 /2
d
c , an
Cl -
_ { 3c/2 if condition (8.1.34) holds

c/2 if condition (8.1.35) holds.
C2 -
Proof. The proof can be modelled on that of Theorem 8.1.2. So, we

shall only give a few details. Let c be a given nonzero number. Without
loss of generality, we may assume that c> O. By conditions (8.1.36) and
(8.1.37) and the fact that Pc(t) -+ 0 as t -+ 00, there is a sufficiently
large T such that for t:::: T the following three conditions are satisfied
I;f(ClS)IQ(s)lds S; ct/4, (c + 2) Itoo f'(c2u)IQ(u)ldu S; 1 and (c + 2)
x I; Pc(s)ds S; c/4. Let F be the Fnkhet space of all continuously
differentiable functions on [T, (0) with the same topology as in the proof
of Theorem 8.1.2, and let X be defined by
X = {x
F : Ix(t) - ctl
S;
(c/2)t, Ix'(t)1
S; c +
f(Clt)Q(t)
+(c + 2)Pc(t) for t:::: T} .

Clearly, X is a nonempty closed convex subset of F. Now, for
consider a mapping fll: X -+ F defined by
(fllx)(t) = ct + [
Q(s)f(x(s))ds + [
t:::: T
(J.oo Q(U)f'(X(U))X'(U)dU) ds.
569
As in the proof of Theorem 8.1.2 it can be shown that (11) cI> is welldefined on X and maps X into itself, (h) cI> is continuous on X, and
(h) cI>X is compact. By the Schauder-Tychonov fixed point theorem the
operator cI> has a fixed point x E X. This fixed point x = x(t) is a
solution of equation (8.1.1) satisfying (8.1.39) and (8.1.40).

Example 8.1.2. Consider the equation (8.1.28). Applying Theorem 8.1.6
to the case f(x) = Ixl" sgn x, IQ(t)l::; 2lklt'\ A < 0, we see that if
A < -" then (8.1.28) has a solution x(t) such that x(t) = et + O(tO) as
t -+ 00 if A+,+l =I 0 where (\ = max{A+,+l,O} and x(t) = cHO(lnt)
as t -+ 00 if A + , + 1 = O.
We note that Theorem 8.1.6 is also 'sharp' in the sense that equation
(8.1.28) in the sublinear case (0 < "( < 1) never has a nonoscillatory
solution unless A < -,.
The following corollaries of Theorem 8.1.6 are immediate.
Corollary 8.1.3.
Suppose either condition (8.1.34) or (8.1.35) holds.
Also, suppose conditions (8.1.36) - (8.1.38) hold. Then for any nonzero
constant c equation (8.1.1) has a solution x(t) such that the asymptotic
condition (8.1.5) holds.
Corollary 8.1.4. Suppose either condition (8.1.34) or (8.1.35) holds. Also
suppose for every k =I 0,
f(kt)Q(t) -+ 0 as t -+
and
J=
f'(ks)IQ(s)lds <
(8.1.41)
00
(8.1.42)
00.
Then for any nonzero constant e equation (8.1.1) has a solution x(t)
such that
x(t) = t[e + 0(1)],
x'(t) = e + 0(1)
as
t -+
00.
(8.1.43)
Next, we shall prove the converse of Corollaries 8.1.3 and 8.1.4 when
Q( t) is of constant sign.
Theorem 8.1. 7 . Suppose Q (t) :2: 0 for all large t. Moreover, suppose
either condition (8.1.34) or (8.1.35) holds, f'{x) > 0 for x =I 0 and for
all nonzero constants kl' k2'
(8.1.44)
Chapter 8
570
Then equation (8.1.1) has a solution
nonzero constant c provided
lim
~ jt If(ks)IQ(s)ds
t-+oo t
JOO f'(ks)Q(s)ds
JOO If(k
<
00
s)IJ'(k2s)Q2(S)ds <
x(t)
satisfying (8.1.5) for some
for some nonzero constant k,
for some nonzero constant k,
00
(8.1.45)
(8.1.46)
for some nonzero constants kl and k 2.

(8.1.47)
Proof. Let x(t) be a solution of equation (8.1.1) which satisfies (8.1.5).

We may assume that c > O. There is a number T such that (c/2)t ~
x(t) ~ 2ct for t:::,: T. As in the proof of Theorem 8.1.3, by employing
Lemma 4.1.3, it follows for t:::,: T that
x'(t)
= af(x(t + Q(t)f(x(t + f(x(t t
00 (
x'(s)
f(x(s
)2 f'(x(sds,
(8.1.48)
where a is a nonnegative constant. On the other hand, an integration by
parts of (8.1.1) from T to t gives
x'(t) = f3 + Q(t)f(x(t - It Q(s)f'(x(sx'(s)ds,
(8.1.49)
where f3 = x'(T) - Q(T)f(x(T. Combining (8.1.48) and (8.1.49), we get
af(x(t)+ f(x(t
1 (f~:~~)
00
f'(x(sds
f3- ltQ(s)f'(x(Sx'(S)dS.
(8.1.50)
Since x'(t):::,: 0 by (8.1.48), (8.1.50) implies
It Q(s)f'(x(sx'(s)ds <
00.
(8.1.51)
From (8.1.48), we get x'(t):::,: Q(t)f(x(t for t:::,: T, and hence we

conclude that
f(x(sf'(x(sQ(s)ds < 00, which in view of (8.1.34)
or (8.1.35) implies (8.1.47). By (8.1.51) we find that the right-hand side of
(8.1.50) has a finite limit as t -t 00. Let () denote that limit, i.e.,
f;

lit
+~ It
x(t) - X(T)
571
T to t and dividing by t, we get
Q(s)f(x(s))ds + -O!
t
(l
f(x(s))
CXJ
It
f(x(s))ds
(f~~~~))
f'(X(U))dU) ds.
(8.1.53)
Since x(t) satisfies (8.1.5), from (8.1.52) and (8.1.53) it follows that
lim -1
t--'tCXJ
It
Q(s)f(x(s))ds = c -
(8.1.54)
f).
Now, it is easy to see that
lit
o ::; t
Q(s)f(x(s))ds
: ; ~ 1J,(~:)
(
ds
) 1/2 (
1
t
f(x(s))f'(x(s))Q2(s)ds
) 1/2
(8.1.55)
h
- - { c/2 if condition (8.1.34) holds
Note that
were c 2c if cOl1dition (8.1.35) holds.
(8.1.44) implies that there exists a positive constant m independent of T
such that
1
f(2cs)
t2
f'(cs) ds ::; m 2 for t ~ T.
for
T,
t
iT
Taking the limit as t -+
o ::;
C -
f) ::;
00
m (
in (8.1.55), we have from (8.1.54),
iT(CXJ f(x(s))f'(x(s))Q2(s)ds ) 1/2
(8.1.56)
Since T is arbitrary, letting T -+ 00 in (8.1.56), we find that c =

Therefore, we have from (8.1.52) and (8.1.54) that
c
and
1
lim -
t--'tCXJ
jt Q(s)f(x(s))ds
O.
f).
(8.1.57)
(8.1.58)
From (8.1.58), (8.1.45) is immediate. By (8.1.48) and (8.1.57), we obtain

x'(t) ~ c/2 for all large t. Combining this with (8.1.51), we see that
(8.1.46) is satisfied. This completes the proof.
572
Chapter 8
For equation (8.1.2), Corollary 8.1.3 and Theorem 8.1.7 yield the following result.
Theorem 8.1.8. Suppose Q(t)
statements are equivalent:
0 for all large t. Then the following
(i)
for any c
satisfying (8.1.5),
i-
0 there exists a solution x(t)
(ii) for some c

satisfying (8.1.5),
i-
0 there exists a solution x(t) of equation (8.1.2)
of equation (8.1.2)
(iii) the following integral conditions are satisfied
/>0 s"i-lQ(s)ds
<
00
(8.1.59)
and
(8.1.60)
Now we are ready to prove the following result.
Theorem 8.1.9. Let 0 < , < 1. Suppose condition (8.1.6) holds and that
Q(t) ~ 0 for t ~ to. Then the following three statements are equivalent:
(i)
(ii)
equation (8.1.2) has a nonoscillatory solution x(t) satisfying (8.1.5),

equation (8.1.2) has a nonoscillatory solution,

Proof. It is trivial that (i) implies (ii). The equivalence of (i) and (iii)
follows from Theorem 8.1.8. Here, we claim that (ii) implies (iii). Theorem
5.1.6 shows that if
s"i-lQ(s)ds = 00, then equation (8.1.2) is oscillatory. This means that if equation (8.1.2) has a nonoscillatory solution, then
the condition (8.1.59) must be satisfied. Therefore, it is enough to show
that if equation (8.1.2) has a nonoscillatory solution, then the condition
(8.1.60) is satisfied.
Joo
Suppose that equation (8.1.2) has a solution x(t), say, x(t) > 0 for
to. Then the equalities and inequalities in the proof of Theorem 8.1.1
remain valid. Now, we consider the following two possibilities:
(h) 2,-1 ~ O. Then the desired condition (8.1.60) follows from (8.1.13)
and (8.1.17).
(12) 2, - 1 > O. In (8.1.15), the first three terms of the right-hand side
are nonnegative and kl (t) is nonincreasing on [to, 00). Thus, we have
x(t) ~ ,(t - to)k1(t) for t ~ to. Hence, in view of (8.1.8), (8.1.11) and
(8.1.14), we get x(t) ~ ,(t-tO)k2(t) for t ~ to, and so by the assumption
573
2,-1> 0,
;::: ,21'-1(t - t o)21'-lQ2(t)
Q2(t)x21'-1(t)k~-21'(t)
Since k~(t) = _Q2(t)x 21'-1(t) for t;::: to,

t ;::: to, gives
- 2! 2, k~-21' (t)
2, k~-21' (to) ;:::
+ 2!
for
t;::: to. (8.1.61)
an integration of (8.1.61) for
,21'-1
1:
(s - to)21'-lQ2(S)ds.
Note that the left-hand side of the above inequality is bounded on [to, 00).
This implies that the desired condition (8.1.60) is satisfied. This completes
the proof.
The equivalence of (ii) and (iii) in Theorem 8.1.9 can be restated as

follows.
Corollary 8.1.5. Let 0 < , < 1. Suppose condition (8.1.6) holds and
that Q(t);::: 0 for t;::: to. Then equation (8.1.2) is oscillatory if and only
if
Theorem 8.1.10. Suppose either Q(t);::: 0 or Q(t):s: 0 for all large

t. Also suppose either condition (8.1.34) or (8.1.35) is satisfied. Then in
order for equation (8.1.1) to have a solution x(t) satisfying (8.1.43) for
some nonzero constant c it is necessary that
lim f(kt)Q(t)
t-'>oo
for some nonzero constant k
(8.1.62)
and
/00 f'(ks)IQ(s)lds
<
00
for some nonzero constant k.
(8.1.63)
Proof. First, we shall prove (8.1.62). If f(x) is bounded as x ---+ 00 or

- 00, then (8.1.62) is trivially satisfied since limHoo Q(t) = O. Thus, we
assume limx-,>oo f(x) = oo. Let x(t) be a solution of equation (8.1.1)
satisfying (8.1.43). By Lemma 4.1.3, we have
Q(t)f(x(t)) = x'(t) - f(x(t))
1 (f~:~~))
00
f'(x(s))ds
for
t;::: T.
(8.1.64)
Using l'Hospital's rule we find that the second term of the right-hand side
of (8.1.64) tends to c as t ---+ 00, i.e.,
d
( dt
itroo (
x' (s ) )
f(x(s))
f (x(s))ds
d
1 )
dt f(x(t))
-1
x (t) ---+
as t ---+
00.
Chapter 8
574
It follows from (8.1.64) and (8.1.43) that limt-HX) Q(t)f(x(t = 0, which

implies (8.1.62). Integrating (8.1.1) by parts, we have (8.1.49). Since
limHOO x'(t) = c and limHOO Q(t)f(x(t = 0, we find that Joo IQ(s)
x 1f'(x(slx'(s)lds < 00 and from this (8.1.63) follows. This completes
the proof.
Combining Corollary 8.1.3 with Theorem 8.1.10, we get the following

result.
Theorem 8.1.11. Suppose either Q(t) ~

or Q(t):S
t. Then the following three statements are equivalent:
for all large
(i)
for any constant c =I=- 0, there exists a solution x(t) of equation
(8.1.2) satisfying (8.1.43),
(ii) for some constant c =I=- 0, there exists a solution x(t) of equation
(8.1.2) satisfying (8.1.43),
(iii) the following integral conditions are satisfied
lim {YQ(t) =
t-+oo
/00
and
s'Y-1IQ(s)lds <
00.
It is to be noted that the condition (iii) of Theorem 8.1.11 can be replaced by limt-+oo s'Y q( s )ds exists and is finite.
8.1.4. Further Extensions and Improvements

The purpose of this subsection is to show that if solutions of (8.1.1)
satisfy (8.1.3) or (8.1.5), then (8.1.6) can be relaxed to the rather weak
hypothesis
litiS
lim sup t-+oo t
to to
q(u)duds > -
00.
(8.1.65)
Further, it will be shown that under the hypothesis (8.1.65) if equation

(8.1.1) has a solution which satisfies either (8.1.3) or (8.1.5), then
lim ~
t-+oo t
it is
to
to
q(u)duds
exists and is finite.
(8.1.66)
Therefore, we may introduce the function Q2(t) by
Q2(t)
lim
T-+oo
lITIS
-T
q(u)duds,
to.
(8.1.67)
The role of Q(t) in the above results is thus taken by the function Q2(t).
In fact, it can be shown that if Q(t) is replaced by Q2(t) in the results
of Subsections 8.1.1 - 8.1.3, then the corresponding results remain valid.
575
Clearly, condition (8.1.6) implies (8.1.65) and the function Q2(t) is

equal to Q(t). Thus, the results we shall present in this subsection generalize those given above.
We begin this subsection with a result on the existence of solutions x
satisfying the asymptotic conditions (8.1.43) and
x(t) = c+o(l),
x'(t)=o(C 1 )
as
t---+oo,
(8.1.68)
where c is a nonzero constant.

Theorem 8.1.12. If equation (8.1.1) has a solution x(t) satisfying either
(8.1.43) or (8.1.68) for some constant c i= 0, then (8.1.6) holds.
Proof. Suppose equation (8.1.1) has a solution x(t) satisfying either
(8.1.43) or (8.1.68) for some constant c i= O. There is a number T 2: to
such that x(t) i= 0 for t 2: T. Dividing both sides of equation (8.1.1) by
f(x(t)) and integrating over [T, t], we find
x'(t)
x'(T)
f(x(t)) - f(x(T))
rt (f(x(s))
x'es) )2 ,
rt
f (x(s))ds+ iT q(s)ds = 0 (8.1.69)
+ iT
for t 2: T. To complete the proof, it is enough to show that

.
x'(t)
11m
--t-400
and
f(x(t))
lOO (f~:~~)))
exists and is finite
J'(x(s))ds <
(8.1.70)
00.
(8.1.71)
When x(t) satisfies (8.1.68) with c i= 0, we have
x'(t)
f(x(t))
-1
= oCt )
and
x'(t) )
( f(x(t))
f (x(t))
and hence (8.1.70) and (8.1.71) are satisfied.

When x(t) satisfies (8.1.43) with c =f. 0, we have x(t) ---+ 00 or
- 00 as t ---+ 00, so by the hypothesis on the function f, limHoo f(x(t))
exists in [-00,00), (i.e., the extended real line). Then, (8.1.70) is trivially
satisfied. Next, note the equality
iT
f'(x(s))x'(s) ds
J2(x(s))
f(x(T)) - f(x(t))'
(8.1.72)
Since the right-hand side of (8.1.72) has a finite limit as t ---+ 00, so does
the left-hand side. Therefore, in view of the fact that x'(t) ---+ c =f. 0
576
Chapter 8
as t ---+
proof.
00,
we conclude that (8.1.71) is satisfied. This completes the
Remark 8.1.3. Theorem 8.1.12 implies that when condition (8.1.6) does
not hold, equation (8.1.1) does not have any asymptotically linear solution x of the form (8.1.43) or (8.1.68). Therefore, the assumption of the
condition (8.1.6) in Subsections 8.1.1 - 8.1.3 is justified.
We now return to the problem of existence of solutions x satisfying
the asymptotic conditions (8.1.3) and (8.1.5). Note that the asymptotic
behavior of the derivatives of solutions x is not restricted in (8.1.3) and
(8.1.5).
Theorem 8.1.13. Assume that the condition (8.1.65) is satisfied. If

equation (8.1.1) has a solution x(t) which satisfies either (8.1.3) or (8.1.5)
for some constant c i- 0, then in addition to the condition (8.1.66) the
equality
(8.1.73)
holds for all large t, where a;::::
is a constant.
Proof. Let x(t) be a solution of equation (8.1.1) which satisfies either

(8.1.3) or (8.1.5) for some constant c i- 0, and let T;:::: to be a number
such that x(t) ifor t;:::: T. Then as in the proof of Theorem 8.1.12
we find that (8.1.69) is satisfied for t;:::: T. Integrating (8.1.69) over [T, t]
and dividing by t, we obtain
r(t) f(u)
du
-
ix(T)
x'(T)
f(x(T))
(t--t T)
rt r (f(X(T))
X'(T) )2 ,
- +t
f (x(T))dTds
1
litis
+t
iT iT
q(T)dTds =
for
t;:::: T. (8.1.74)
The first term of the left-hand side of (8.1.74) has a finite limit as t ---+ 00.
In fact, for the case when x(t) satisfies (8.1.3) with c i- 0, it is clear
that this term tends to zero as t ---+ 00. For the case when x(t) satisfies
(8.1.5) with c i- 0, we assume that c>
(the same will be done for the
Icase c < 0), then it follows that
1r(t) [ 1 1]
t~~ t
ix(T)
f(u) - f(oo) du
and consequently, the same term in (8.1.74) tends to c/f(oo) , where

f(oo) = limf(x) as x ---+ 00. Denote by a the nonnegative limit of the
first term of the left-hand side of (8.1. 74).

The limit as t -+
of the third term of the left-hand side of (8.1.74) is
00
{(Xl ( x' (r) )
iT
f(x(r))
(8.l.75)
f (x(r))dr,
which is finite or infinite. Taking the upper limit as t -+

we find
x'(T)
a- f(x(T))
577
00
in (8.1.74),
{=( f(x(r))
x'(r) )2
t t
f (x(r))dr+h~~~p t iT iT q(r)drds =
+ iT
O.
From this equality and condition (8.1.65) it follows that the integral (8.1.75)
is finite. Then in the limit as t -+ 00 in (8.1.74), we find that (8.1.66)
holds, and moreover, we have
0: -
x'(T)
f(x(T))
{= ( x'(r)
+ iT
f(x(r))
)2 f (x(r))dr + Q2(T)
I
0,
(8.l.76)
where Q2(t) is defined in (8.1.67). Since equality (8.l.76) holds as long

as the number T is chosen such that x(t) # 0 for t 2: T, we may
regard T as arbitrary provided T is chosen sufficiently large. Thus, we
see that (8.1. 73) is satisfied for all sufficiently large t. This completes the
proof.
A further improvement of Theorem 8.1.13 can be obtained by relaxing

(8.l.66) to the weaker condition: there exists an integer n 2: 1 such that
lim
t--+oo
11
-tn -
it
to
(t - s)"-1q(s)ds
exists and is finite.
(8.l.77)
Then, we can define the function Qn as
Qn(t) =
iT
1
lim --=1
T-+cxo
Tn
(T - s)n-1q(s)ds,
(8.1.78)
which coincides with the function Q when n = 1 and with Q2 for

n = 2. Obviously, each one of the hypotheses (8.l.6) and (8.1.66) implies
(8.1. 77).
Now we shall prove that if equation (8.1.1) has a solution x(t) satisfying
(8.1.3) or (8.1.5) for some constant colO and there exists an integer n 2: 1
such that
limsup
t--+=
n~1
it
then condition (8.1.77) holds.
to
(t - s)n-1q(s)ds > --
00,
(8.1.79)
Chapter 8
578
Theorem 8.1.14. Let n 2': 2 and let condition (8.1.79) be satisfied. If

equation (8.1.1) has a solution x(t) which satisfies either condition (8.1.3)
or (8.1.5) for some constant c -I- 0, then (8.1.77) holds, and
x'(t)
f(x(t))
roo (
+ Qn(t) + it
x'(s)
f(x(s))
)2 ,
f (x(s))ds
(8.1.80)
for all large t, where a is a nonnegative constant.

Proof. The proof for the case n = 2 is given in Theorem 8.1.13. So, we
shall assume that n 2': 3. Let x(t) be a solution of equation (8.1.1) defined
for t 2': To for some To 2': to and satisfies (8.1.3) or (8.1.5), where c is
a nonzero constant. From (8.1.3) or (8.1.5) it follows that x(t) -I- 0 for all
large t. Clearly, there is no loss of generality in assuming that x( t) -I- 0
for all t 2': To
From equation (8.1.1), for every T, t with T 2': t 2': To, we have
So, for T 2': t 2': To it follows that

1
Tn-l
itr (T T
s)
n-l
1T
1
- (n - l)(n - 2) Tn-l
1
- Tn-l
1T(
T-s
q(s)ds
)n-l (
( t )n-l ( x'(t) )
1- T
f(x(t))
(T - s)n-3
X'(S)
f(x(s))
[1
X'(T) dT ] ds
f(X(T))
(8.1.81)
)2 f '( x ()) ds.

s
Next, we shall show that the

1
T~ Tn-l
itrT(T - s )n-3 [1
X'(T) d ] d
f(X(T)) T S
(8.1.82)
exists as a nonnegative real number, and it is independent of t (2': To).

Denote this limit by 6. For this, we assume that the point t 2': To is fixed.
We first observe that the limit in (8.1.82) can be written as
1
lim Tn-l
T .....
oo
1T
t
(T - s)n-3
[l
(S)
x(t)
-du
f () ds.
U
(8.1.83)
579
If the solution x(t) satisfies condition (8.1.3) for some constant c #- 0,

then x(t) is bounded and hence J:(~~) du/ f(u) is also bounded for all
s 2: t. In this case it is clear that the above limit equals zero. It remains to
consider the case when x(t) satisfies condition (8.1.5) for some constant
c #- O. Assume that c > O. (When c < 0, the proof is similar and
hence omitted). This means that x(t) is positive on [To, =). Consider an
arbitrary number c> 0 with 0 < c < c. The limit f(=) = lim" .... oo f(u)
exists as a positive number or infinity. So, we can choose a Uo > 0, so
that (l/f(u)) - (l/f(oo)) < c for all u 2: Uo. Furthermore, by taking
into account (8.1.5), we can conclude that there exists a tl > t with
(c - c)tl 2: Uo and such that
(c-c)r ::::: x(r) ::::: (c+c)r
forevery
r2:t 1 .
(8.1.84)
Then for any T, s with T 2: s 2: t 1 , we obtain
1 1 1]
X
(S) [
x(t)
f(u) - f(oo) du
1 1] I + 11x(t,)(S) [ f(u)1 - f(oo)1] du I

11 (t') [_1 __
1 ]dul + l(c-<)t,
r(c+<)s cdu
x(t)
f(u)
f(oo)
< 11 (t') [
x(t)
<
f(u) - f(oo) du
11;t~t') [ftu) - f(~)] dul + c[(c + c)T -
<
(c - C)tlJ.
Thus, for every T 2: tl it follows that
T:-I
liT(T-s)n-3{1;t~S) [ftu) - f(~)] dU}dSI
< T:-l iT (T -
<
s)n-311;t~S) [f/U) - f(~)] dul ds
{Il;t~t') [f/U) - f(~)]dUI+c[(C+c)T-(C-C)tlJ}

x
and consequently,
(n! (1 _~ )
2)T
n-2 ,
Chapter 8
580
Since
is arbitrary, we find
)~!,
(T - s)n-3
{lX(S) [f(u)
1x(t)
By using (8.1.84), we obtain for T 2': t l
T
T:_Ii (T-S)n-3
I]}
f(oo) du
ds
o.
(8.1.85)
[rx{s) dU] ds
.Jx(t)
T:-I itl (T - s)n-3 x (s)ds

t
+ T:-I
iT (T - s)n-3x (s)ds
t,
1
1 (
t ) n-2
x(t)
- (n - 2) T 1 - T
<
;2 rtl
~
T
X(S)dS+(C+f)T:_Ii (T-s)n-3 sds
t,
__1_~
(1- .!.)n-2 x(t)
2)
itl x(s)dS+(C+f) [(n-2)Ttl ( tlT )n-2
T2
(n -
I-
(tl)n-l]
1
1 (
t )n-2
1- T
- (n _ 2) T 1 - T
x(t)
- (n - 2)(n - 1)
C+f
-+ (n-2 )( n-1 ) as T -+
and
TLI iT (T - st- 3
[l
(S)
00
dU] ds
x(t)
> -1
T2
l
1 iT
1- t
~) n-3 i t x(s)ds + (c - f ) (T - s)n-3 sds
T
t
Tn-I t,
_ _1_~
(n - 2) T
-..!...
T2
(1 _.!.)n-2 x(t)
(1- tlT )n-3 i

1
+ (n - 2)(n - 1)
T
tl
(tl)n-l]
T
1-
C-f
-+ (n-2 )( n-l )
x(s)ds + (c _ f)
as
T -+
00.
[_I_tl
(1- tlT )n-2
(n - 2) T
1
1 (
(n _ 2) T
t)n-2
1- T
x(t)

Therefore, since
1
lim Tn-1
T--+oo
581
is arbitrary, we conclude that
[T (T - s)n-3 [lX(S) du1ds

. x(t)
= ( _ 2)( _ )"
n
(8.1.86)
From (8.1.85) and (8.1.86) it follows that the limit (8.1.83) is equal to the
real number
J =
(n - 2)(n - 1)
(1)
f(oo)
and this proves our assertion.

Now, from (8.1.81), we have
lim sup T:-1 (T(T - s)n-1q(s)ds
lTD
T--+oo
x'(To)
.
1
{T(
)n-1 ( x'(s)
f(x(To)) - a - )~~ Tn-1 lTD T - s
f(x(s))
)2 f '( x ( )) ds,
s
where a = (n -l)(n - 2)8. Thus, condition (8.1.79) guarantees that

.
{T(
)~~ Tn-1 lTD T - s
)n-l ( x'(s)
f(x(s))
)2 f '(x (s ))ds
<
00.
Hence, for all t;::: To,
T~m= Tn-1
[T(T
t
00
)2 f'( x (s ))ds
x'(s) )2 ,
f(x(s))
f (x(s))ds <
)n-1 ( x'(s)
f(x(s))
S
t
00.
Finally, from (8.1.81) we conclude that condition (8.1.77) is satisfied, and

lim T
T--+oo
1
n
-1
[T (T - s)n-1q(s)ds
[= (f(x(s))
x'(s) )2 ,
f (x(s))ds
t
x'(t)
f(x(t)) - a -
for
t;::: To.
So, the solution x(t) satisfies (8.1.80). This completes the proof.
In view of Theorem 8.1.14 it is reasonable to restrict our attention to

the case when condition (8.1.77) holds in discussing the existence of asymptotically linear solutions x(t) of the form (8.1.3) and (8.1.5). Now, it is
important to note the following lemma.
Lemma 8.1.1. Let
n;::: 2 be an integer. Then
582
Chapter 8
(i) if (8.1.77) holds, then the function Qn is continuously differentiable

on [to,oo) and Q~(t) = -q(t), moreover
1
lim
-1
tn -
t-->oo
it
to
(t - s)n- 2Qn(s)ds = 0,
(8.1.87)
(ii) if the function h(t) is continuously differentiable on [to, 00) such

that h'(t) = -q(t) and
lim
t-->oo
--1
tn -
it
to
(t - s)n-2h(s)ds
then condition (8.1.77) holds and Qn(t)

For any T, t with T
Proof. (i)
1
Tn-l
Jtr
iT
iT
to
- (n - 1)
However, since for T
IT:-11:(T- s
T:-1
1:
lim TL1
T-->oo
lim
T-->oo
1
n-1
1T
t
(T - s)n-1q(s)ds -
T
~
t- 2
:-1 it
to
[1:
(T - s)n-2
it follows that
Thus, for all t
to, we have
it
to
(T - s)n-2
Tt)n-l1tto q(r)dr
[1
q(T)dr] ds.
to
to,
q(r)dr] dsl
[1:
(T - s)n-2
(T_s)n-lq(s)ds
i.e.,
(T - s)n-1q(s)ds
1-
Iq(r)ldr] ds <
[1
to
Qn(tO)
.!.It [1
T
q(r)dr] ds
to,
Qn(t)
it
1
(T - s)n-1q(s)ds - Tn-1
Tn-1
(8.1.88)
= h(t).
(T - s)n-1q(s)ds
1
Tn-1
0,
1
lim Tn-1
T-->oo
-it
to
to
for
iT
to
q(r)dr
to
Iq(r)ldr] ds
t ~ to.
(T-s)n-lq(s)ds-
for
~ to.
it
to
q(r)dr,
(8.1.89)
583
This ensures that Qn is continuously differentiable on [to, CXJ) and satisfies

Q~(t) = -q(t). Furthermore, for every t:::: to,
t n1- 1
1t
(t - st- 2 Qn(s)ds
l:(t -
to
tn~l
st- 2 [Qn(to) -
L
S
q(T)dT] ds
1
(
t )n-1
1
(n _ 1) 1 - :
Qn(tO) - tn-1
t)
rt (t -
st- 2
.flo
1
-1- ( 1-..!!. n-1 Q (to) - -1- (n - 1)
t
n
(n - 1) t n - 1
11
to
[1 q(T)dT
S
to
ds
(t - st-1q(s)ds
from which (8.1.87) follows immediately in view of the definition of Qn.

(ii)
From h'(t)
= -q(t), it follows that

h(t) = h(to)
-1t
q(s)ds
for
t:::: to.
(8.l.90)
to
So, we can derive for every

for Qn)
t n 1- 1
t::::
(t - s)n-2h(s)ds
to (by using the same arguments as above
to
1
(
t
- - 1-..!!.
(n - 1)
t
- -(-1
- ) - -11
n - 1 tn -
)n-I h(to)
1t (
t - s )n-l q (s )ds.
to
Thus, in view of (8.1.88) we conclude that. (8.1.77) holds, and Qn(tO) =

h(t o). Finally, from (8.1.89) and (8.1.90) we find that Qn(t) = h(t). This
completes the proof of the lemma.
Remark 8.1.4. It can be easily verified that the results presented in

Subsections 8.1.1 - 8.1.3 remain valid with the function Qn(t) (where
n:::: 1 is an int.eger) in place of Q(t), and cOlldition (8.1.77) instead of
(8.1.6).
To illustrate the above results, we discuss the following example.
Example 8.1.3.
q(t) = {
Consider equation (8.1.2) with
nA sin (nA(t - 2mr))
otherWIse,
where A > 0 is a constant and
for
t E In,
n = 1,2,
(8.1.91)
Chapter 8
584
Here q(t) is regarded as a function on the interval lRo
h(t)
+ cos(nA(t ~ 2n1T
t E In, n
for
[0, (0). Let
1,2,
(8.1.92)
otherwise.
Then the function h is continuously differentiable on lRo and h'(t) =

~q(t) for t 2: O. Since h(t) has no limit as t -> 00, Theorem 8.1.12
implies that equation (8.1.2) with (8.1.91) has no solutions x of the form
(8.1.43) and (8.1.68). Noting that 0 s: h(t) s: 2 for t E In, n = 1,2, ...
we find that if t E In for some n = 1,2, ... , then
lit
8
n
8
n
o s: -t o
h(s)ds n
s: (2 ~ n -A)"'"'
1T
Ii ds = 2n - n- A "'"'iA .
Thus, limHoo(1/t) J~ h(s)ds = 0 and so, by Lemma 8.1.1, (8.1.88) with
n = 2 is satisfied, and Q2(t) = h(t). Noting that Q2(t) (= h(t
satisfies Q2(t) s: 2 for t E In, n = 1,2, and Q2(t) 2: 1 for
t E I n , n = 1,2, ... , where
we can easily show that integral conditions (8.1.21), (8.1.22), (8.1.59) and
(8.1.60) with Q2(t) inplace of Q(t), are satisfied if and only if ). >
1, ). > 2, ). >, and ). > 2" respectively. Therefore, we can conclude by
Theorem 8.1.2 with Q2(t) inplace of Q(t) and condition (8.1.66) instead
of (8.1.6) that equation (8.1.2) with (8.1.92) has a solution x(t) of the
form (8.1.3) (c =1= 0) if and only if ). > 2. Similarly, we can conclude by
Theorem 8.1.8 with Q2(t) inplace of Q(t) and condition (8.1.66) instead
of (8.1.6) that equation (8.1.2) with (8.1.91) has a solution x(t) of the
form (8.1.5) (c =1= 0) if and only if ). > 2,.
8.2. Asymptotic Behavior of Positive

Solutions of Singular EmdenFowler-Type Equations
In this section we shall discuss the positive solutions of the singular
Emden-Fowler-type equations
x"(t)
+ q(t)x-A(t)
x"(t)
= q(t)x-A(t)
0,
(8.2.1)
(8.2.2)
585
and the system
xl/(t)
p(t)y-A(t)
q(t)x-J.L(t) ,
y"(t)
(8.2.3)
where p(t), q(t) E C([to,oo),JR) and A, f.l are positive constants.

In Subsections 8.2.1 and 8.2.2 we shall investigate the problem of existence and asymptotic behavior of positive solutions of equations (8.2.1)
and (8.2.2). Subsection 8.2.3 is concerned with the positive solutions of the
system (8.2.3).
8.2.1. Positive Solutions of Equation (8.2.1)

If x(t) is a solution of (8.2.1) which exists and is positive for all sufficiently large t, then we call x(t) a proper solution of equation (8.2.1).
If x(t) is a proper solution of (8.2.1), then xl/(t):s; 0 and since x(t) > 0
we find that x'(t) decreases to nonnegative limit as t -+ 00. Therefore, if x(t) is a positive solution of (8.2.1) then either: (I) x(t) is a
proper solution in which case either (II) 0 < limt-HXl x'(t) < 00 or (b)
limt-too x'(t) = 0, or (II) x(t) is not a proper solution and there exists
b < 00 such that limt-tbx(t) = 0 and x(t) cannot be continued after b
as a twice continuously differentiable solution of (8.2.1).
In what follows we are concerned with the proper solutions of equation (8.2.1) and discuss existence, uniqueness and asymptotic behavior of
solutions satisfying (II) or (1 2 ).
Existence and Uniqueness of Positive Solutions

Let a > 0 and c be fixed real numbers. In what follows by x(t, (3)
we shall denote the solution of
xl/(t)
+ q(t)x-A(t)
Theorem 8.2.1. If
00
0,
s-Aq(s)ds =
x(c)
00,
= a, x'(c)
/3.
to;:=: 1
(8.2.4)
to;:=: 1
(8.2.5)
to
then for all
/3, x(t, /3)
falls into case (II). If
00
s-Aq(s)ds <
00,
to
then there exists

(al) if
/3 < /30,
/30
such that
then x( t, /3) falls into case (II),
586
Chapter 8
(a2) x(t, /30) satisfies (1 2),

(a3) if /3 > /30, then x(t, /3) satisfies (II)'
Furthermore, limHOO x(t,{3o)
< 00 if and only if
Ita; sq(s)ds < 00.
Theorem 8.2.1 is a direct consequence of the following lemma and two

propositions.
Lemma 8.2.1. Suppose a < b, q(t), ql(t) E C([a,b],lR o), q(t) < ql(t)
and x(t) and yet) are positive solutions on [a,b] of equations (8.2.1)
and
y"(t) +ql(t)y-A(t) = 0
respectively. If x(a) 2': yea) and x'(a) > y'(a),
x'(t) > y'(t) for t E (a,b].
Proof. For t E [a, b], we have
x(t) = x(a)
and
yet)
yea)
-it
-it
+ (t - a)x'(a)
+ (t - a)y'(a)
then x(t) > yet) and
(t - s)q(s)x-A(s)ds
(t - s)ql(s)y-A(s)ds.
Subtracting these equations, we get
x(t) - yet)
[x(a) - yea)]
+
it
+ [x'(a) - y'(a)](t - a)
(t - s) [ql(S)y-A(s) - q(S)X-A(S)] ds.
(8.2.6)
Since x'(a) > y'(a), there exists an t > 0 such that x(t) > yet) for
a < t :::; a + t. Suppose x(t) :::; yet) for some t E (a, b]. Then there exists
e E (a, b] such that x(c) = y(e) and x(t) > yet) for a < t < c. Letting
t = e in (8.2.6), then the left-hand side is zero and the right-hand side is
positive, which is a contradiction. Hence, x(t) > yet) for a < t :::; b, and
since
x'(t) = x'(a)
-it
we have x'(t)
> y'(t) for a:::; t
q(s)x-A(s)ds and y'(t) = y'(a)

:::; b.
-it
ql(s)y-A(s)ds,
Proposition 8.2.1. Let c be a real number. Then there exist solutions

of (8.2.1) which are positive for t 2': c if and only if condition (8.2.5)
holds. Furthermore, if (t > 0 and condition (8.2.5) holds, then there exist
solutions of
x"(t) + q(t)x-A(t) = 0, x(c) = (t
(8.2.7)

which are positive for t:::: c and limt--+oo :r' (t) >
587
o.
Proof. Let x(t) be a solution of equation (8.2.1) which is positive for

t :::: c. Let z( t) be the solution of
-z"(t) = q(t)[x(c)
+ x'(e)(t - c)]->-,
z(e) = :1:(c),
z'(e) = x'(e).
Since 0 <S x(t) <S :r;(c)+x'(e)(t-e) for t:::: e, we have -z"(t) <S -y"(t)
for t:::: c. Hence, z'(t) - :J;'(t) = f:[z"(,5) - ;r;"(s)]ds > 0 and since
:J:'(t) :::: 0 for t:::: c, we have z'(t):::: 0 for t:::: c. But,
Z'(t)
and therefore,
x'(e)
00
-it
q(s)[x(e)
q(s)[x(c)
+ x'(e)(s -
+ x'(c)(s -
ell-Ads <
ell-Ads
00.
Hence, flOG s-Aq(s)ds < 00. Conversely, suppose that condition (8.2.5)
holds and a> O. Choose (3 > 1 + feoo q(s)[a + s - cl-Ads. Let x(t) be
the solution of equation (8.2.7) with x'(e) = (3. If x'(t) > 1 for all t :::: e,
then clearly x(t) is positive and limHoo x'(t) > O. If :J;'(t) <S 1 for some
t> c, let a = inf{t > c: x'(t) = I}. Then, y'(t):::: 1 for c <S t <S a and
hence x(t)::::a+(t-c) for e<St<Sa. So,
(3
= x'(c) = x'(a) +
q(s)x-A(s)ds
and this contradicts the choice of (3.
l
+1
<
1+
<
00
q(s)[a
+ s - erAds
q(s)[a + s - cl-Ads
Proposition 8.2.2. Let a > 0 and c be real numbers. If condition

(8.2.5) holds, then there exists a unique solution x(t) of
x"(t)
+ q(t)x-A(t) =
Furthermore, x( 00)
0,
x(e) = a,
x'(oo) = O.
(8.2.8)
< 00 if and only if
00
sq( s )ds <
00.
(8.2.9)
Proof. Without any loss of generality we can assume that e = O. Suppose

x(t) and y(t) are two positive solutions of (8.2.8) with y'(O) > x'(O).
Then by Lemma 8.2.1, y(t) > x(t) for t > O. Hence,
(y(t) - x(t))" = q(t) [x-A(t) - y-A(t)] > O.
Chapter 8
588
So, y'(t) - x'(t) 2:: y'(O) - x'(O) for t 2:: 0, and hence y'(oo) - x'(oo) 2::
y'(O) - x'(O). This contradicts y'(oo) = x'(oo) = 0, and hence the
uniqueness follows.
Let
x{3 (t)
be the solution of
x"(t) + q(t)x->'(t) = 0,
x(O) =
x'(O) = {3.
0:,
Let T = {{3 : x~(t) 2:: 0 for t 2:: O} and S = {(3 : x~(t) < 0 for
some t 2:: O}. By Lemma 8.2.1 if {31 E Sand {32 E T, then {31 < {32.
By Proposition 8.2.1, T i= (/) and every negative number belongs to S.
Thus, sup S = inf T and if b = sup S, then 0::; b < 00. By continuous
dependence on initial conditions b tJ. S. So, bET and we need only to
show that x~(oo) = O.
Let x(t) = Xb(t) and suppose x'(oo) > O. Then, x(t)j(t + 1) is
continuous and positive for t 2:: 0 and limHoo x(t)j(t + 1) = x'(oo) > O.
Therefore, if M = inf {x(t)j(t + 1) : t 2:: O}, then M> O. Choose 8> 0
such that
roo q(s)(s + 1)->'ds)
2 ] >.+1
8 exp ( >. [ M
Jo
M
<""4
and b - 8> Mj2 (since b 2:: x'(oo) 2:: M this is possible). Let y(t) =
Xb-t5(t). If y(t)j(t + 1) > Mj2 for all t > 0, then clearly b - 8 E T
contradicting b = infT. Suppose y(t)j(t + 1) ::; M/2 for some t > O.
Since y(O) = x(O) > M/2, y(t)/(t + 1) 2:: M/2 for 0::; t ::; a, where
a = inf{t > 0: y(t)j(t + 1) ::; Mj2}. Also, y(a)j(a + 1) = Mj2. Now, for
o ::; t ::; a, we have
Now, by Gronwall's inequality, we have for 0::; t ::; a,
~ _ .!!l!l
t +1
and hence
+1
::; 8 exp
x(a)
a+1
[>. (2)
>'+1 t
M
Jo
< y(a) + M
a+1
q( s)
1<
(s + 1)>' ds
3M
4
4'
589
contradicting the choice of M. Thus, x' (00) = 0 and this proves the
existence.
Let x(t) be the solution of (8.2.8). Then integrating (8.2.8) from t
to u and letting u -+ 00, we obtain
x'(t) =
00
Integration of (8.2.10) from 0 to

x(oo)
q(s)x-A(s)ds.
00
= a+
00
(8.2.10)
now gives
sq(s)x-A(s)ds
(8.2.11)
and since a < x(t) < x(oo) it follows from (8.2.11) that x(oo) <
and only if condition (8.2.9) holds.
00
if
Asymptotic Behavior
The following theorem deals with the asymptotic behavior of proper
solutions of equation (8.2.1) which satisfy case (Id.
Theorem 8.2.2.
Let x(t) be a proper solution of equation (8.2.1)
satisfying (II). Then there exists a constant a > 0 such that
(8.2.12)
Furthermore, if
00
s-Mlq(s)ds <
00,
to::::: 1
(8.2.13)
to
then there exists a constant b > 0 such that
x(t) = at + b - a- A[l + 0(1)]1 (s - t)s-Aq(s)ds

00
as t -+
00.
(8.2.14)
Proof. Let a = limHoo x'(t). Then, a> 0 and x(t) = at[1 + 0(1)] as
00. Integrating equation (8.2.1) from t to v and letting v -+ 00,
we obtain
t -+
x'(t) - a
00
q(s)x-A(s)ds
= a- A[1 + 0(1)]1 s-Aq(s)ds. (8.2.15)

00
Integrating (8.2.15) from to::::: 1 to t, we find (8.2.12).

Suppose (8.2.13) holds. Then the right-hand side of (8.2.15) is integrable on [1, (0) and hence limt-too[x(t) - at] exists. Denote this limit
Chapter 8
590
by b. Then integrating (8.2.15) from t to
x(t)
= at+b-a->'[l+o(l)Jl
OO
00
100
gives
u->'q(u)duds.
Interchanging the order of integration in (8.2.16), we get (8.2.14).
(8.2.16)
Next, we present the asymptotic behavior of the proper solutions of

equation (8.2.1) satisfying (12) and condition (8.2.9).
Theorem 8.2.3.
Let x(t) be a proper solution of equation (8.2.1)
satisfying (12). If condition (8.2.9) holds, then there exists a constant a > 0
such that
x(t) = a - a->'[l
+ o(l)J 100 (s - t)q(s)ds
as t --+
00.
(8.2.17)
Proof.
Since condition (8.2.9) holds, we have by Theorem 8.2.1 that
limt-400 x(t) = a < 00 for some constant a > O. Integrating equation
(8.2.1) from t to u and letting u --+ 00, we obtain
x'(t) =
100 q(s)x-A(s)ds
Integrating (8.2.18) from t to
x(t)
a - a- A[l
a->'[1 + 0(1)]100 q(s)ds.
=
00
gives
+ o(l)J 100
00
q(u)duds.
Interchanging the order of integration in (8.2.19), we get (8.2.17).
(8.2.18)
(8.2.19)
To study the asymptotic behavior of proper solutions of equation (8.2.1)

satisfying (1 2 ) and the condition
00
sq(s)ds
00,
(8.2.20)
we need the following three lemmas.

Lemma 8.2.2. Let c be a positive constant. Then, x(t) is a solution
of equation (8.2.1) if and only if C1 /(A+l)X(t) is a solution of the equation
y"(t) + cq(t)y-A(t) = O.
Proof. The verification is obvious, and hence left to the reader.
Lemma 8.2.3. Let y(t) and w(t) be two positive proper solutions of
equation (8.2.1) satisfying (h). If condition (8.2.20) holds, then y(t) rv
z(t) as t --+ 00. (The notation y(t) rv z(t) as t --+ 00 means that
limHOO y(t)/z(t) = 1).
591
Proof. Without any loss of generality, we can assume that yet) and z(t)
are defined for t::::: 0 and yeO) > z(O). By Proposition 8.2.1, condition
(8.2.5) holds. And, by Proposition 8.2.2, yet) cannot intersect with z(t).
Thus, yet) > z(t) for t::::: O. Also, y'(t):s; z'(t) for t::::: 0, if for
y'(a) > z'(a) for some constant a::::: 0, then for t::::: a,
y'(t)-z'(t) = y'(a)-z'(a)+
tq (S) [z->'(s) - y->'(s)] ds::::: y'(a)-z'(a) > 0,
contradicting limHoo y'(t) = limHoo z'(t) = O. Hence, (y(t) - z(t))

decreases to a nonnegative limit as t ---+ 00. Therefore, for t::::: 0,
yet)
yet) - z(t)
yeO) - z(O)
1 :s; z(t) = 1 +
z(t)
:s; 1 +
z(t)
.
(8.2.21)
Since condition (8.2.20) holds, we have by Proposition 8.2.2 that z(t) ---+
as t ---+ 00. Thus, by (8.2.21), yet) '" z(t) as t ---+ 00.
,>
00
Lemma 8.2.4. Suppose q(t), ql(t) E C([a,oo),R o), a::::: 0, q(t):s; ql(t)
and
0 is a constant. Let x(t) and yet) be positive solutions of
equation (8.2.1) with x(a)
x'(oo) = 0 and the equation
="
(8.2.23)
respectively. Then, y(t)::::: x(t) for t::::: a.
Proof. Suppose there exists a constant b > a such that y(b) < x(b).
Let c = max{t < b : yet) ::::: x(t)}. Then, y(e) = x(e) and yet) < x(t)
for e < t :s; b. Thus, for some dE (e, b), y'(d) < x'(d) and y(d) < xed).
Hence, by Lemma 8.2.1, yet) < x(t) for t::::: d. Therefore,
x"(t) - y"(t) = ql(t)y->'(t) - q(t)x->'(t) >
for
t::::: d.
Hence, x' (t) - y' (t) ::::: x' (d) - y' (d) > 0 for t::::: d, contradicting
limHoo x'(t) = limHoo y'(t) = O. Thus, y(t)::::: x(t) for t::::: a.
Theorem 8.2.4. Suppose q(t), ql(t) E C(Ro, R+), limHoo ql(t)/q(t) =

r > 0, where r is a constant, and x(t) and yet) are positive solutions
of equation (8.2.1) with x'(oo) = 0, and
y"(t)
+ ql(t)y->'(t)
0,
y'(oo) = 0,
respectively. If (8.2.20) holds, then yet) "-' rl/(>'+l)x(t) as t ---+
00.
Proof. Let pet) = ql(t)/q(t), t ::::: 0 and the numbers E, 'rl E (0,1).
Choose a constant a> 0 such that (1- E)r < pet) < (1 + E)r for t::::: a.
We can assume that x(t) and yet) are defined for t::::: O. By Proposition
Chapter 8
592
8.2.1, condition (8.2.5) holds. And, by Proposition 8.2.2 there exist unique
solutions YI (t) and Y2(t) of
y~'(t)
+ (1- E)rq(t)Y1A(t)
0,
YI(a)
y~(t)
+ (1 + E)rq(t)Y2"A(t) =
0,
Y2(a) = yea), Y~(=) = 0,
= yea),
y~(=) = 0
and
respectively. Since (1 - E)rq (t) ::; qI (t) ::; (1 + E)rq (t) for t 2: a, we have
by Lemma 8.2.4 that YI(t)::; yet) ::; Y2(t) for t;::: 0,. By Lemma 8.2.2,
[(1- E)rJI/(A+Ilx(t) and [(1 + E)rj1/(A+ Il x(t) are solutions of
y~(t)
+ (1- E)rq(t)Y1A(t) =
0,
y~(=)
0,
y~(=)=o,
and
y~(t)+(I+E)rq(t)Y1A(t)
=0
respectively. By Lemma 8.2.3, [(1 - E)rp/(A+Ilx(t) '" YI(t) and [(1

E)rj1/(A+ Il x(t) '" Y2(t). Hence, there exists b 2: 0, such that for t 2: b,
Since E and 'f/ are arbitrary numbers in (0,1), yet) '" rl/(A+I)x(t) as
t~=.
q(t)

q(t) rv ql(t) as t ~ =,
00
C(JRo,JR+), ql(t) E CI(JRo,JR+),
Sql (s)ds
(8.2.23)
and x(t) is a positive solution of equation (8.2.1) with x'(=) = o.

Let J(t) = [Jtoo s-Aql(s)ds]I/(A+Il. If limHootf"(t)/f'(t) = r where
r > -2 is a constant, then x(t) rv ((A + 1)/(2 + r))I/(A+Il(tJ(t)) as
t~=.
Proof. By Proposition 8.2.1 condition (8.2.5) holds. Since q(t) rv ql(t) as

~
condition (8.2.23) is also satisfied. Hence, J(t) is well-defined and
by Proposition 8.2.2 there exists a solution yet) of y"(t)+ql(t)y-A(t) = 0,
y'(=) = O. Let z(t) = tJ(t), t 2: o. Then,
z"(t)
+ B(t)Z-A(t) =
0,
where
1 [
tf"(t)]
B(t) = A + 1 2 + J'(t) ql(t).
(8.2.24)
593
Since z(t) > 0 and B(t) > 0 for all sufficiently large t, we have by (8.2.24)
that limHoo z'(t) ~ O. Let 00= limHOO z'(t). If a> 0, then z(t) ~ oo(t)
as t ---c> 00, and hence f(t) ~ a as t ---c> 00, contradicting f(t) ---c> 0 as
t ---c> 00. Hence, z'(oo) = O. Also, limHoo B(t)/ql(t) = (2+r)/(1+'\) > O.
So, by Theorem 8.2.4,
z(t)
rv
(12+ 'r) 1/(.\+1) y(t)

+,\
(21 + r)
1/(.\+1)
+,\
x(t)
as
---c> 00.
Therefore,
x(t)
1+,\)1/(.\+1)
( -tf(t)
2+r
as
---c> 00.
Remark 8.2.1.
1. In Theorem 8.2.5, the number r can be calculated by the following
equation
tJ"(t)
(8.2.25)
I'(t)
2. In Theorem 8.2.4, the asymptotic behavior of proper solutions of (8.2.1)

depends on the asymptotic behavior of q(t). Using Theorem 8.2.4, we
can find ill Theorem 8.2.5 the asymptotic behavior of proper solutions of
equation (8.2.1) under rather mild conditions on the function q(t).
As an application of Theorems 8.2.1 - 8.2.5, we shall determine the
asymptotic behavior of all proper solutions of equation (8.2.1) when q(t) ~
ql(t) as t ---c> 00, where ql(t) = tQ(lnt);3, and a and (3 are constants.
One of the following possibilities must hold:
(ir) 0 < 00+2 < ,\ + 1,
(i2) 0 < 00+2 = ,\ + 1 and (3 > -1,

(h) 0 = 00+2 and (3 > -1,
(i4) 0 = 00+2 and (3 = -1,
(i5) 00+2<0 or(oo+2=0 and (3<-1),
(i6) 00+2 > ,\ + 1 or (a + 2
= ,\
+ 1 and (3
-1).
ftC; s-.\q(s)ds < 00 and

holds if and only if ftC; sq(s)ds <
only if ftC; s-.\q(s)ds = 00, to ~ 1.
Each case (h) - (i4) holds if and only if
ftC; sq(s)ds =
00,
00, to ~ 1. Case (i5)

to ~ 1, and case (i6) holds if and
By Proposition 8.2.1, there is no proper solution of equation (8.2.1) if

(i6) holds.
594
Chapter 8
If (i5) holds and x(t) is a proper solution of equation (8.2.1), then by

Theorems 8.2.1 and 8.2.2 there exist constants a and b, a > 0 such that
(8.2.26)
as t --+
or
00,
x(t) = a-a->-[1+0(1)lloo(S-t)so(Ins)!3ds
(8.2.27)
By using the fact that

t Tn + 1
---(In
-m-1
(In t)n+l
-n -1
tt
if
if
< -1
(8.2.28)
=c=
-1
and
n <-1
the integrals in (8.2.26) and (8.2.27) can be simplified.

Suppose (h) or (i2) holds. Let f(t) =
using (8.2.28) and (8.2.25), we get
r=
tf"(t)
hm--=
Hoc
f'(t)
(!too s->-q1(s)ds) 1/(>-+1) .
Then
a +2
..
, - - 2 If (ld holds
{ ,,+1
-l'f
1 (.12 ) 110ld S.
Since r > -2, we have by Theorem 8.2.5 that if x(t) is a proper solution
of equation (8.2.1) satisfying (12)' then
x(t) ~ (
A+
r
1)
+2
1/(>-+1)
(A
tf(t)
+ 1)2
(a+2)(A-1-a)
] 1/(>.+1)
-(3 - 1
] 1/(>-+1(lnt)!3/(>-+1)t(2+c/(>.+1) if (i ) holds
t(ln t)(!3+1)/(>.+1) if (i 2) holds.
If x(t) is a proper solution satisfying (i1)' then by Theorem 8.2.1 there

exists a > 0 such that
x(t)
at+o(l:lO u O ->-(lnU)!3dudS)
as
t--+oo.
If (i3) or (i4) holds, we cannot use Theorem 8.2.5 because r = -2. If
(b) holds, let
y(t) = (
~: ~ )
1/(>-+1)
(In t)(!3+ 1 )/(>-+l).
595
Then, - y"(t)y>'(t) '" q1(t) '" q(t) as t -+ 00. Hence, by Theorem 8.2.4,
if x( t) is a proper solution of equation (8.2.1) satisfying (i3) then
(8.2.29)
for some constants a and b, a >
o.
If (i4) holds, we let y(t) = (>.+ l)l/(>.+l)(1nlnt)l/(>.+l). Then, _y"(t)

x y>'(t) '" ql (t) '" q(t) as t -+ 00. Hence, by Theorem 8.2.4, if x(t)
is a proper solution of equation (8.2.1) satisfying (i 4 ), then x(t) '" (>. +
1) 1/(>.+1) (In In t) 1/(>.+1) .
8.2.2. Positive Decaying Solutions of Equation (8.2.2)

By a positive decaying solution of equation (8.2.2) we mean a function
E C 2 ([to, (0), lR+), to 2:: 0 which solves equation (8.2.2) and satisfies
the asymptotic condition
x(t)
lim x(t) = 0
lim x'(t) = O.
and
t ..... oo
t ..... oo
(8.2.30)
In what follows we shall assume that

(i) the support of q(t) is not compact, and
(ii) the improper integral
Q(t)
00
q(s)ds
(8.2.31)
converges for all t 2:: to and Q(t) 2:: 0 on [to, (0).
Existence and Uniqueness of Positive

Decaying Solutions
We shall prove two existence theorems by means of a variant of the
shooting method; one applies to the case when q(t) E C([to, (0), lR), and
the other when q(t) E C([to, (0), lRo). For this, we shall need the following
basic lemma.
Lemma 8.2.5. Suppose the function Q(t) satisfies
[00 Q(s)ds <
ito
00
(8.2.32)
and
[00 sQ2(s)ds <
ito
00.
(8.2.33)
596
Chapter 8
Then for any constant P> 0 equation (8.2.2) has a positive solution x(t)
satisfying
(8.2.34)
lim x(t) = P,
lim x'(t) = 0
t--">oo
t--">oo
and x'(t) < 0 on [to, 00).

Proof. Clearly, x(t) E C 2([to, 00), ffi+) is a positive solution of equation
(8.2.2) satisfying (8.2.34) if and only if x(t) solves the integral equation
In view of (ii) this equation is equivalent to the integro-differential equation
x(t)
Q(s)x- A(s)ds->.lO(lOQ(u)x-A-1(u)X'(U)dU) ds, t
00
P+
~ to.
(8.2.35)
First, we shall solve equation (8.2.35) in some neighborhood of infinity,
say, for t ~ T, T being sufficiently large. As in Theorem 8.1.1, choose
k, m > 0 and T ~ to so that
p-A
oo
Q(s)ds + >.p-A-l m
+>.P-A-1k
P- A ::::; k
and
oo
sQ2(s)ds
(h
OO
(h
OO
Q(S)dS)
>.p-A-l ( m
+k
oo
SQ2(S)dS)
< P,
Q(S)dS) ::::; k.
With this choice of k, m and T consider the closed convex subset X

of the Fnkhet space C1([T,00),ffi) (equipped with the usual topology)
consisting of all functions x(t) satisfying P::::; x(t) ::::; 2, and
o ::; For t
x(t) ::::; mQ(t)
+k
00
Q2(s)ds
for
~ T.
T and each x E X, we define the mapping F: X -+ X by
(Fx)(t) = P+
00
Q(s)x- A(s)ds->.lO
(1
00
Q(u)x-A-1(u)X'(U)dU) ds.
Then it is easy to verify that F maps X continuously into a compact

subset of X. Therefore, the Schauder-Tychonov theorem implies that F
has a fixed point x in X, i.e., x(t) = (Fx)(t) for t ~ T. Clearly, this
x(t) gives a positive solution of the equation (8.2.35) on [T,oo).
597
We claim that x'(t) < 0 for t 2': T. To this end suppose that X'(T)
for some T E [T, 00 ). Since
=0
x'(t) = - Q(t)x-A(t) + A
.(JO Q(s)x-A-I(s)x'(s)ds
for
t 2': T (8.2.36)
from the nonpositivity of each term on the right-hand side of (8.2.36), we

find that
Q(s)x-A-l(s)x'(s)ds = 00, which implies that Q(t)x'(t) ==
o on [T, (0). Multiplying (8.2.36) by Q(t), we get Q(t) == 0 on [T, (0),
i.e., q(t) == 0 on [T, (0). This contradicts the assumption (i). Thus, we
have x'(t) < 0 for t 2': T.
I;
Now, we can continue x(t) to the left as a solution of equation (8.2.2).

Let J c [to, (0) be the maximal interval of existence for x( t). Clearly,
x(t) satisfies (8.2.35) and (8.2.36) for t E J. We claim that x'(t) < 0
throughout J. In fact, if this is not true, we can find T E J, T < T such
that X'(T) = 0, x'(t) < 0 in (T, (0). Then by putting t = T in (8.2.35)
we arrive at a contradiction
o=
Q(T)X-A(T)
+A
LX) Q(s)x-A-I(s)x'(s)ds
< O.
Thus, x'(t) < 0 for t E J.

Finally, we shall show that J
= [to, (0).

there exists a
#- [to, (0). Since x(t) > 0 and x'(t) < 0 on J,
tl E [to, (0) such that
x(t) -+
as
00
Then, X-ACt) is bounded on (tl' T),

equation
x(t)
= x(T) + x'(T)(t - T) + [,
t -+ tt.
(8.2.37)
and this fact combined with the
(is q(U)x-A(u)dU) ds,
(tl, T]
which follows from (8.2.2) shows that x(t) tends to a finite limit as t -+ tt.
This contradicts (8.2.37), and so, we must have J = [to, (0), i.e., x(t)
exists on the whole interval [to, (0). This completes the proof.
Theorem 8.2.6. Suppose condition (8.2.32) holds. If
100
to
slq(s)1
(100 Q(u)du)
-A/(A+1)
ds <
00,
(8.2.38)
then equation (8.2.2) has a positive decaying solution.
Proof. We note that (8.2.38) implies

(8.2.33) is satisfied. For each n E lN,
ItC; slq(s)lds < 00
so that condition
let xn(t) be a positive solution
Chapter 8
598
of equation (8.2.2) satisfying x~(t) < 0 for t::::O: to, and xn(t)---t
lin, x~(t) ---t 0 as t ---t 00. The existence of such Xn (for each n) is
guaranteed by Lemma 8.2.5. Also, xn(t) for t::::O: to satisfies
Xn(t)
=
Since
1 (1
~ +1
~+
00
00
x~(t)
00
q(U)x;;A(U)dU) ds
Q(s)x;;A(s)ds - A
1 (1
00
00
Q(u)x;;A(u)X'(U)dU) ds.
(8.2.39)
< 0 for t::::o: to, from (8.2.39), we obta.in
which implies
00
Xn(t)::::O:
Q(s)ds
l/(Ml)
(8.2.40)
t::::O: to
From (8.2.40) and the first equality of (8.2.39) it follows that
xn(t)
1
+1
<
~+
<
(s - t)lq(s)1
(1
slq(s)1
Q(U)dU) -A/(A+l) ds,
00
00
(1
00
00
Q(U)dU) -A/(MI) ds
t::::O: to.
(8.2.41 )
Similarly,
x~(t)
Ix~(t)1 ~
can be estimated as follows
00
Iq(s)1
roo
is
Q(u)du
-A/(A+l)
ds,
t::::o: to
(8.2.42)
Inequalities (8.2.41) and (8.2.42) assert that the sequence {xn(t)} is uniformly bounded and euicontinuous on each compact subset of [to, 00).
Therefore, Ascoli-Arzela's theorem implies that {xn} has a subsequence
{XnJ converging to some x E C([to,oo),lR) uniformly on each compact
subset of [to, 00). From (8.2.40) it follows that x(t) > 0 for t::::o: to
Letting ni ---t 00 in the equation
we find by Lebesgue dominated convergence theorem that x(t) satisfies
599
Hence, x(t) is a positive decaying solution of equation (8.2.2). This completes the proof.
Example 8.2.1. Consider equation (8.2.2) with
Since q(t)
00
q(t) =
_(I~sint)/,
t +c
(C 2 -
as t --t
Q(s)ds
f )
t-l-f
1+
E>A.
and
+ 0 (C 2 -
--
00,
t;:::l,
f )
;::: CC 1 -
as
t --t
00,
for some constant C > 0, a.ll conditions of Theorem 8.2.6 are sa.tisfied and
hence there exists a positive decaying solution of this equation.
Theorem 8.2.7. Let q(t);::: 0 for t;::: to and A E (0,1). Suppose there
exists a nonincreasing function p( t) E C ([to, 00 ), lR+) such that
s:
q(t)
and
00
pet)
t;::: to
for
p1/2(s)ds <
(8.2.43)
(8.2.44)
00.
to
Then equation (8.2.2) has a positive decaying solution.
J::
Proof. We notice that (8.2.44) implies

sp(s)ds < 00, which in turn
implies conditions (8.2.32) and (8.2.33). Let {xn(t)} C C([to,oo),lR+)
denote the same sequence as in the proof of Theorem 8.2.6. Now, we shall
show that the following estimates hold
X};'+1)/2(t) <
and
(-1)
(.H1)/2
(X~(t))2 s:
(1
and noting that x~(t)

for t;::: to,
1/2
00
1 2
p / (s)ds
t > to
,-
(8.2.45)
C~ A) p(t)x~->'(t),
Integrating equation (8.2.2) (with x
(~[x~(t)f)'
2)
+ -+2-A) (I-A
t;::: to
(8.2.46)
= x n ), we find
q(t)x;;->'(t)x~(t)
for
t;::: to
< 0 on [to,oo) and limHoo x~(t) = 0, we obtain
Chapter 8
600
This establishes (8.2.46). Since (8.2.46) can be rewritten as
_(X~1+>')/2(t))'
::;
C;>') C~>.y/2p1/2(t)
for
t"2to
an integration of this yields (8.2.45).
By (8.2.45), {xn} is uniformly bounded on each compact subset of

[to, (0) and so is {x~} by (8.2.46). Moreover, equation (8.2.2) with
x = Xn together with (8.2.40) shows the uniform bounded ness and the
equicontinuity of {x~} on each compact subset of [to, (0). Hence, it
follows by Ascoli-Arzela's theorem that there exists a subsequence {xni}
of {xn} which converges to a function x E C 2 ([to,00),lR) in the C 2 topology on each compact subset of [to, (0). It is clear that x(t) > 0 on
[to, (0). Letting ni ---* 00 in the equation X~i(t) = q(t)X;;i>'(t), t"2 to
and the inequality
X~:+1)/2(t):::;
( ni1)(1+>')/2 + (1 >.) ( 1-2 >. )1/2100

.
pl/2(S)ds,
t
t"2 to
we see that x(t) is the desired positive decaying solution of equation

Remark 8.2.2.
When ft'; Q(s)ds = 00, equation (8.2.2) has no positive decaying

solution regardless of the sign of the function q(t). To show this, let x(t)
be a positive decaying solution of (8.2.2). Multiplying (8.2.2) by x>'(t)
and integrating from t to u and letting u ---* 00, we get
1.
-X'(t)X>'(t) -
>.100 x>.-l(s)(x'(s)?ds
= Q(t),
t"2 to
which implies - (x>'+l(t)j(>. + I)' "2 Q(t) for t"2 to. Integrating this
inequality from t"2 to to T"2 t, we obtain
Letting
---+ 00, we find that

_l_x>'+l(t) "2
>'+1
100
Q(s)ds,
t"2 to.
(8.2.47)
Therefore, the integrability condition on Q(t) is natural for the equation

(8.2.2) to have a positive decaying solution.
2. The above results can be extended easily to more general equations of
the form
(a(t)x'(t))' + 8q(t)x->'(t) = 0,
601
Joo
where 8 = 1, a(t) E C([to, 00), JR+) and

dsja(s) = 00, by changing
the independent variable s =
du/a(u). The details are left to the reader
(this transformation has been already used in the previous chapter).
It:
Next, we present the following uniqueness theorem for positive decaying

solutions of equation (8.2.2).
Theorem 8.2.8. Let q(t) 2: 0 for t 2: to. Suppose condition (8.2.32)
holds and that the equation
w'(t)
A 1 q(t)
+ A+
(1
t
00
Q(s)ds
)-1 w(t) =
0,
(8.2.48)
is nonoscillatory. Then the positive decaying solution of (8.2.2) is unique.

Proof. Let Xl (t) and X2(t) be positive decaying solutions of equation
(8.2.2). Suppose Xl (t) 2: X2(t) in some neighborhood of infinity, say, for
t 2: tl 2: to Since Xi(t) 2: JtOC! (fsoo q(u)x;A(u)du) ds, t 2: to, i = 1,2 it
follows that Xl(t) ::; X2(t) for t 2: tl and consequently that Xl(t) == X2(t)
for t 2: tl. Since the uniqueness of solutions of the initial value problem
for equation (8.2.2) ensures that Xl(t) == X2(t) on [to, tIL we conclude
that Xl(t) == X2(t) on [to, 00).
From the above observation it suffices to show that the difference w(t) =
X2(t) - Xl(t) cannot change sign infinitely often on [to, 00). Suppose the
contrary. Then, w(t) is an oscillatory solution of the linear differential
equation
w'(t)
+ c(t)w(t)
0,
t 2: to
(8.2.49)
where
Since each x
that
= Xi
satisfies (8.2.47), it follows from the mean value theorem
A 1 q(t)
c(t) ::; >. +
(1
t
00
Q(s)ds
)-1
for t 2: to.
In view of the nonoscillatory character of (8.2.48), Sturm's comparison

theorem implies that equation (8.2.49) must be nonoscillatory, which is
a contradiction. This completes the proof.
The following corollary is an immediate consequence of Theorem 8.2.8

and the well-known Hille's theorem.
Chapter 8
602
Corollary 8.2.1.
If
Let q(t) 2: 0 for t 2: to and condition (8.2.32) hold.
A 1 t 2q(t)
A+
(1
00
Q(s)ds
)-1
::;
for all large
t,
then the positive decaying solution of equation (8.2.2) is unique.

Example 8.2.2. Consider (8.2.2) under the condition that c l r 2 -<> ::; q(t)
::; C2r2-o., t 2: 1 for some constants Ci > 0, i = 1,2 and a > O. By
Theorem 8.2.6 and Corollary 8.2.1, if
then equation (8.2.2) has exactly one positive decaying solution. In fact,
the singular Emden-Fowler equation x"(t) = r 2-o.x-)..(t), t 2: 1, a> 0
has exactly one positive decaying solution
x(t) =
provided Aa(a
) 2 ] l/()"+l)
A+ 1
a(a + A + 1)
Co./()"+l)
for
t 2: 1
+ l)/(A + 1) ::; 1/4.
Asymptotic Behavior of Positive Decaying Solutions

Consider (8.2.2) and the equation
y"(t)
p(t)y-)..(t) ,
(8.2.50)
where p(t), q(t) E C1([to, 00), JR.+). In the following theorem, we assert
that if p(t) and q(t) have the same asymptotic behavior in some sense,
then so do the positive decaying solutions of these equations as t ---+ 00.
For this, we notice that when equation (8.2.50) admits a positive decaying
solution, then the integral P(t) = ftoo(s - t)p(s)ds converges for t 2: to.
To prove our result we shall need the following elementary lemma.
Lemma 8.2.6. If x(t) E Cl([to, 00), JR.) satisfies Joo x2(s)ds < 00 and
Ix'(t)l::; c, t 2: to for some constant c> 0, then limHoox(t) = O.
Theorem 8.2.9. Let x(t) and y(t) be positive decaying solutions of
equations (8.2.2) and (8.2.50), respectively. Suppose that
lim q(t) = 1,
t--+oo
lim sup p-l/()..+l)(t)

t--+oo
00
p{t)
(s - t)p(s)P-)../()..+l)ds < 00,
(8.2.51 )
(8.2.52)
o
where Y(t)
< liminf Y(t)

t-HXJ
lim sup Y(t) < 00,
603
(8.2.53)
t-+oo
[(p(t)y3->'(t))-1/2]', and either
00
p1/2(s)y-(1+>'l/2(s) Iq(s) p(s)
11 ds
< 00,
(8.2.54)
or
(8.2.55)
is satisfied. Then, x(t),..., y(t) as t ---+ 00.
Proof. The hypotheses of the theorem imply that for t
C1 P1 /(>.+l l (t)
x(t),
y(t)
~ c21O (s -
to,
t)p(s)P->'/('\+ll(s)ds (8.2.56)
for some constants Ci = Ci(A) > 0, i = 1,2. In fact, the first inequality
in (8.2.56) is obtained from (8.2.47), while the second inequality in (8.2.56)
follows from y(t) = ftoo (fsoo p(u)y->'(u)du) ds, t ~ to. Define the new
function v by v(t) = x(t)/y(t) for t ~ to. Then, v(t) satisfies the
equation
vl/(t)
+ 2 (~(~?) v'(t) + p(t)y->'(t)v(t) =
q(t)y->.-l(t)v->'(t) (8.2.57)
for t ~ to. Introducing the new independent variable T =

equation (8.2.57) transforms into
d2 v
dT2
+ p(t)y3->'(t)v =
q(t)y3->'(t)v->-,
ftto
y-2(u)du,
T ~ O.
(8.2.58)
Moreover, since Iooo (p[t(T)]y3->'[t(T)j) 1/2 dT = 00 by (8.2.53), the change

of variable s
into
v-
= I; (p[t(~)ly3->' [t(~)]) 1/2 d~

f(s)v
+v =
g(s)v->-,
transforms equation (8.2.58)
~ 0,
(-
= :s)
(8.2.59)
where
f(s) =
[ (p(s)y3->'(s) -1/2]' y2(S)
and
g(s) = q(s)/p(s)
for
s ~ O.
Notice that (8.2.51), (8.2.53) - (8.2.55) are equivalent to the conditions

lim g(s) = 1,
8-+00
(8.2.60)
Chapter 8
604
o<
liminf f(s) S; lim sup f(s) <

8-;CX)
s--+oo
/00 Ig(s) _ lids

and
/00 Ig(s)lds
<
<
00,
(8.2.61 )
(8.2.62)
00
(8.2.63)
00,
respectively. It follows from (8.2.52) and (8.2.56) that

m S; v(s) S; M,
s2:0
(8.2.64)
for some constants m, M> O. Condition (8.2.61) implies that

(8.2.65)
for So sufficiently large and some constants kl' k2 > O. For simplicity,
we suppose that So = 0 and ).. i- 1. The proof will be complete if we
can show that lims->oo v(s) = 1. From (8.2.60) and (8.2.64) there is a
constant K > 0 such that Ig(s)v--X(s) - v(s)1 S; K for s::::: So. Let L
be a constant satisfying L > K/k 1 . We claim that v(s) is bounded in
lRo, i.e.,
Iv(s)1 S; L for all large s.
(8.2.66)
For this, suppose that (8.2.66) fails to hold. Since clearly (8.2.64) implies
liminfs->oo Iv(s)1 = 0, we can find a sequence of intervals {[an, bn]}, n E
IN such that bn < an+l, limn->oo an = 00, Iv(an)1 = IV(bn)1 = L
Iv(s)1 > L, an < s < bn . Then Rolle's theorem shows that there exists
Cn E (an, bn)
such that v(c n) = 0, Iv(cn)1 > L. However, by setting
s = C n in (8.2.59) we arrive at a contradiction for all large n, namely,
klL :s; f(cn)lv(cn)1 = Ig(cn)v--X(cn)1 :s; K. Hence, (8.2.66) holds.
Note that by the hypotheses of the theorem and boundedness of v( s)
it follows that v( s) is also bounded on lRo.
First suppose that condition (8.2.54) holds.
(8.2.59) as
By rewriting equation
and integrating over [0, s], we find
(8.2.67)

where
C3
605
is a constant. Since (8.2.62), (8.2.64) and (8.2.66) give
for some constant

I(X) f(s)i,2(S)ds <
C4
00,
> 0,
(8.2.66) together with (8.2.64) implies that

and hence by (8.2.65), we obtain
Joo v (s)d8
2
<
(8.2.68)
00.
From the boundedness of v(s), (8.2.68) and Lemma 8.2.6, we find that
limHooi'(8) = O. Accordingly, it follows from (8.2.67) that the limit
must exist as a finite value, i.e., lim s-+ oo V(8) = (finite). Letting 8 ---* 00
inequation(8.2.59),weget lims-+oov(8)=->--. If ->-_1-0, then
the boundedness of V(8) is violated. Hence, ->- = (I, i.e., (I = 1 as
desired.
Next, let condition (8.2.55) hold. We notice that equation (8.2.67) is
equivalent to
(8.2.69)
for some real constant C5. Condition (8.2.63) ensures the convergence of the
right-hand side of (8.2.59). Hence, we have (8.2.68) and lims-+cxo {;(8) = 0,
which in turn imply that the limit
exists in JR. The rest of the proof is similar to that of presented above and
hence omitted. This completes the proof.
Remark 8.2.3. As a simple consequence of the proof of Theorem 8.2.7,

we observe that if 0 < A < 1 and there exists a non-increasing function
p*(t) E C([to, (0), JR+) such that p(t) :S; p*(t) for t 2: to, then (8.2.52)
can be replaced by the condition
limsup p-l/2(t)
t-+cxo
cxo
(P*(8))1/2ds <
00.
Chapter 8
606
Consider the case when the function pet) behaves like eta where c
and 0 are constants, c> 0 and 0 < -2, i.e., assume that
< liminf pet) < lim sup pet) <

ta
t .... oo
and
p'(t)
o<
ta
t .... oo
-p'(t)
lim inf ---- < lim sup - - <

ta -
t .... oo
ta -
t .... oo
(8.2.70)
00
(8.2.71 )
00.
Then for any positive decaying solution yet) of equation (8.2.50), (8.2.56)
shows that C 1t(a+2)/(A+l) :::; yet) :::; C 2 t(a+2)/(A+1), t ~ to for some
positive constants C 1 and C 2 . Moreover, the equation - y'(t) =
ftOO p(s)y-A(s)ds, t ~ to gives the estimates C3t(a+2)/(M1)-1 :::; -y'(t)
:::; C4 t(a+2)/(A+1)-1, t
to.
Now, the following corollary is immediate.

Corollary 8.2.2. Let A:::; 3, 0 < -2 and let x(t) and yet) be positive
decaying solutions of equations (8.2.2) and (8.2.50), respectively. Suppose
that conditions (8.2.51), (8.2.70), (8.2.71) and either
/
OOl- l q (S)
- - 1 ds
s pes)
are satisfied. Then, x(t)
rv
<
00,
or
y(t) as t --+
00.

Corollary 8.2.3. Let 0 < -2 and let q(t) = ct a [1 + E(t)], where c is
a positive constant and E(t) E C([to,oo),1R) satisfies limt->ooE(t) = 0,
and either
IE~)I ds < 00,
(8.2.72)
/00
or
/00 1E'(s)lds <
(8.2.73)
00
holds. Then any positive decaying solution x(t) of equation (8.2.2) satisfies
x( t)
Proof.
rv
(0
(A
1)2
+ 2)(0 -
A + 1)
11/(HA)
t(o:+2)/(Ml)
as
--+
00.
It suffices to note that the singular Emden-Fowler equation

c > 0 and 0 < -2 has a positive
y"(t) = ctay-A(t), t ~ to with

decaying solution yet) given by
yet) =
] 1/(A+1)
( A 1)2
+
t(a+2)/(M1)
(0 + 2)(0 - A + 1)
c
for
t>
-
to

for which (8.2.53) (with p(t) = cta) is satisfied.
607
Next, we shall employ the transformation given in the proof of Theorem

8.2.9 to establish the following uniqueness theorem of positive decaying
A:::: 3
Theorem 8.2.10. Let

satisfies
0 <
and
0 <
and a < -2. Suppose the function q(t)
.
q(t)
rllllm
. f -q( t) < IllllSUP
<
t ..... oo
1. .....
to:
t'"
00
00
.
-q'(t)
rImln
. f-q'(t)
- < lUll sup - - 1 - <
1. .....
to<-1
t ..... oo
to:00
00.
Then equation (8.2.2) has a unique positive decaying solution.

Proof. Suppose to the contrary that we have two distinct positive decaying
solutions x(t) and yet) of equation (8.2.2). If the difference x(t) - yet)
is of one sign for all sufficiently large t, then the simple observation we
had in Theorem 8.2.8 implies that x(t) yet). Therefore, we need to treat
the case when x(t) - yet) changes sign infinitely many times. ,As in the
proof of Theorem 8.2.9, we define the new functions vet) = x(t)/y(t) and
I(t) =
y-2(u)du, t ~ to. Then calculations give
It:
d2 v
d/ 2
+ q(t)y3-A(t)[V -
Moreover, the change of variable s =

forms equation (8.2.74) into
v-
f(s)v
+v
v-A,
V-A] = 0,
~ O.
(8.2.74)
I; (q[t(E)]y3-A[t(E)]) 1/2 dE
s
~ 0,
(- = :s)
trans-
(8.2.75)
f(s) = ( (p(S)y3-A(S)) -1/2)' y2(s), S ~ O. Now, in view of the

proofs of Theorem 8.2.9 and Corollary 8.2.1, we have lim s ..... oo v(s) = O.
But, this leads to a contradiction. To see this, let Sn, n E 1N be the point
at which v(s) cuts the horizontal line v = 1 and Sn t 00 as n -* 00.
The existence of such points is guaranteed by the fact that v( s) oscillates
around v = 1. Then an integration of equation (8.2.75) multiplied by v(s)
where
over
[Sn, Sn+1]
yields
which contradicts the fact that limHoo v(s) = 0 unless v(s)

=1.
This
Chapter 8
608
In the following uniqueness theorem, we shall employ Corollary 8.2.3.

Theorem 8.2.11. Let a < -2. Suppose the function q(t) is of the form
q(t) = ct a [l + E(t)], where c is a positive constant, E(t) E C([to, 00), lR),
limHoo tE'(t) = 0 and either condition (8.2.72) or (8.2.73) holds. Then
equation (8.2.2) has a unique positive decaying solution.
Proof. Let x(t) and yet) be two positive decaying solutions of equation
(8.2.2). We may suppose that x(t) - yet) changes sign infinitely many
times as before. We notice from Corollary 8.2.3 and I'Hospital's rule that
x(t) and yet) satisfy
x(t), yet)
rv
KtlT
x'(t), y'(t)
and
rv
(JKt lT - 1 as t ---+ 00, (8.2.76)
where
K =
c().
+ 1)2
] 1/(>+1)
and
(a+2)(a-).+1)
a+2
(J= ),+1'
Then as in the proof of Theorem 8.2.10 we obtain equation (8.2.75). By

taking into account (8.2.76) it is easily seen that
lim f(s)
8--+00
C- 1 / 2 K(A+l)/2[). -
2a
+ 3](). + 1)-1 >
and now the proof can be completed as that of Theorem 8.2.10.
Asymptotic Behavior of Positive Increasing Solutions

This subsection address the asymptotic behavior of positive increasing
solutions of equation (8.2.2) when q(t) E C([to,oo),lR+). If x(t) is a
positive solution of equation (8.2.2), then xl/(t) > 0 eventually, and hence
x'(t) is eventually of one sign. Therefore, we have either (i) x'(t) < 0
eventually, or (ii) x'(t) > 0 eventually. The case (i) when limHoo x(t) =
0, i.e., decaying solutions of equation (8.2.2) is already studied in the
previous subsection. The other case, i.e., case (ii) is our objective here.
A positive increasing solution of equation (8.2.2) is defined to be x(t) E
C 2 ([to, 00), lR+) having the asymptotic behavior
lim x'(t)
t--+oo
lim x(t)
t--+oo
c,
c E lR*
= lR+ U {oo}.
(8.2.77)
Remark 8.2.4. Suppose q(t) satisfies the condition
/00 s-Aq(s)ds < 00.
(8.2.78)
Then any positive increasing solution of (8.2.2) satisfies limt--+oo x'(t) =

limHoo x(t)/t = Cl < 00, where Cl is a constant. For this, since x'(t) >
609
0, t :2: to it follows that x(t):2: kt, t :2: to for some constant k
o.
>
Integrating (8.2.2) from to to t and using x(t):2: kt, t:2: to we obtain
x'(t) = x'(to)
+ it q(s)x-:>"(s)ds ~
x'(to)
+ k-:>..i t s-:>"q(s)ds <
00
for t:2: to. Since x'(t) is an increasing function for t:2: to it must have
a finite limit as t -+ 00, and this completes the proof of the assertion.
From this observation, we restrict our attention to the case when condition (8.2.78) fails to hold. We shall use the following notation: For
t :2: to, Ql(t) = It: s-:>"q(s)ds, Q2(t) = Itto Qi/(:>"+l) (s)ds and Q3(t) =
ItDO Q:;2(s)ds. Now, we shall prove the following lemma.
(8.2.79)
Then any positive increasing solution x(t) of equation (8.2.2) satisfies
. f x(t)
1Imln
Q ()
2
t-+oo
>
(\ + 1)1/(:>"+1).
_ /\
Let x(t) be a positive increasing solution of equation (8.2.2).

Since x" (t) > 0 for t:2: to, it follows that
Proof.
x(t) = x(to) + it x'(s)ds

to
x(t o) + (t - to)x'(t)
for
t:2: to
+ 0(1)] as t -+ 00, which is equivalent to

(x'(t)):>"x"(t) :2: r:>"q(t) [1 + 0(1)] as t -+ 00, or
Hence, (x'(t)):>":2: r:>"x:>"(t) [1
cx~t~:+l)'
:2: C>'q(t)[l
+ 0(1)]
as
t -+ 00.
Integrating the above inequality twice from to to t, we get
and
x(t) :2: (,\ + 1)1/(>.+1)Q2(t)[1
+ 0(1)]
as
t -+ 00.
Lemma 8.2.8. Let the functions J(t), g(t) E C 1 ([to,00),JR) be defined

near infinity such that g'(t) i= 0 and limHoo g(t) = 00. Then,
J'(t)
liminf - t-+oo g'(t)
..
J(t)
J(t)
f'(t)
< hmmf
-() < hmsup -() < hmsup ~().
t-+oo 9 t
t-+oo 9 t
t-+oo 9 t
Chapter 8
610
Remark 8.2.5. If condition (8.2.79) holds, then the function

CQ2(t), t 2: to is a positive increasing solution of the equation
y(t) =
(8.2.80)
where C is a positive constant. Hence, it is natural to expect that the
positive increasing solutions of (8.2.2) behave like y(t) when the coefficient
of (8.2.80) is asymptotic to the function q(t). This observation leads to
the following results.
Theorem 8.2.12. Let 0
< ).. < 1. Suppose (8.2.79) and

Q2(t)
lim
t-+oo tQ~/(>'+l)(t)
a E 1R+
(8.2.81 )
hold. Then any positive increasing solution x(t) of equation (8.2.2) satisfies
(8.2.82)
Theorem 8.2.13. Suppose conditions (8.2.79) and (8.2.81) hold. Further,
assume that
lim q(t)Q~->'(t)Q~(t) = bE 1R+
(8.2.83)
t->oo
and either
00
/
Q3-1( S)Q2-2()
S
Q2(S)
1/(>.+1)
sQ 1
(s)
a ds <
00,
(8.2.84)
or
hold. Then any positive increasing solution x(t) of equation (8.2.2) has
the asymptotic form (8.2.82).
Theorem 8.2.14.
Suppose in addition to (8.2.79), (8.2.81), (8.2.83),
(8.2.84) the following condition holds
Then any positive increasing solution

asymptotic form (8.2.82).
x(t) of equation (8.2.2) has the
Also, from Lemmas 8.2.7 and 8.2.8 the following result is immediate.
611
Theorem 8.2.15. Let 0 < A < 1, and x(t) and y(t) be positive
increasing solutions of equations (8.2.2) (I,nd (8.2.50), respectively, such that
x(t)
;::0:
my(t)
for
t;::o: to,
(8.2.85)
for some constant m > O. Suppose further that condition (8.2.79) and
. q( t)
11m -()
= 1
P t
(8.2.86)
t-+=
hold. Then,
x(t)~y(t)
as t--+oo.
Since the function z(t)

equation
= (A + 1)1/('\+1)Q2(t) for t;::o: to solves the
],\_,\
Q2(t)
z"(t) = q(t) [ tQ~/('\+l\t) z (t)
for
t;::o: tl > to
;::0:
Lemma 8.2.7 and Theorem 8.2.15 show that the assumption 0 < A < 1
in Theorem 8.2.12 is superfluous if a = 1 in condition (8.2.81). We state
this result in the following.
Corollary 8.2.4. If conditions (8.2.79) and (8.2.81) with a = 1 hold,
then any positive increasing solution x(t) of equation (8.2.2) satisfies
x(t) ~ (A + 1)1/(A+l)Q2(t) as t --+ 00.
Theorem 8.2.16. Let x(t) and y(t) be positive increasing solutions
of equations (8.2.2) and (8.2.50) respectively, satisfying (8.2.85). Suppose
conditions (8.2.79), (8.2.86),
lim p(t)y3-'\(t)
t-+cxo
(1= y-2(S)d8)2
t
= E R+,
(8.2.87)
(8.2.88)
and either
(8.2.89)
or
(8.2.90)
hold. Then, x(t)
y(t) as t --+
00.
612
Chapter 8
Proof. From Theorem 8.2.15 it follows that
my(t) :s; x(t) :s; My(t)
for
t 2: to
(8.2.91)
for some constant M > O. Set v(t) = x(t)/y(t) for t 2: to. Since v(t)
satisfies equation (8.2.57), the change of variable T = (JtOO y-2(u)du)-1
implies that v( T) satisfies
for some TO > O. Furthermore, introducing the new independent variable

s = ln T, this equation is reduced to
v + il + p(s)v
q(s)v->"
for
s 2: So = ln TO,
C :s)
=
(8.2.92)
where p(s) = p(t)y3->"(t)T- 2 and q(s) = q(t)y3->"(t)T- 2 for s 2: So

Now, conditions (8.2.86) - (8.2.90) become
lim q(s) = 1,
p( s)
(8.2.93)
= f,
(8.2.94)
8-+00
lim q(s)
1
1 I~~:~ -11
8-+00
00
\(q(s'\ds <
00
(8.2.95)
00
ds <
(8.2.96)
00
and
(8.2.97)
respectively. By (8.2.91), we have
m :s; v(s) :s; M,
s 2: So
(8.2.98)
Since (8.2.93) and (8.2.94) imply that lims-+oo q(s) = f, the argument used
in the proof of Theorem 8.2.9 shows that v(s) is bounded on [so, 00),
and from equation (8.2.92), we see that v(s) is also bounded. Multiplying
equation (8.2.92) by v(s), we get
p-()
S (q(s)
p(s)
-1) v
->...v,
s>
_ so
(8.2.99)
613
We shall prove that lims--+oo v(s) = 1 when condition (8.2.89) holds. The
other case when (8.2.90) holds can be proved similarly, and will not be
included. Integrating equation (8.2.99) over [so, sJ, we obtain
{)2(s)
2
1
-1
+
i,2(7])d7] + p(s)
[v 2(S) _ V1-A(S)]
2
So
(p(7]))' [1'2(7]) _ 'l.,l-A(7])] d7]

2
1 - .\
1:
So
1 - .\
[:~~~
p(rl)
- 1] v- A(7])'v(rl) d71 for
(8.2.100)
s
~ so,
where r is a real constant. Conditions (8.2.95) and (8.2.98) ensure the

convergence of the second integral on the left-hand side of (8.2.100) as
s --+ 00. Similarly, conditions (8.2.94) and (8.2.96) and the boundedness
of v( s) imply the convergence of the integral on the right-hand side, from
which we have Joo v2(s)ds < 00. Hence, lims--+oo v(s) = 0 by Lemma
8.2.6. The rest of the proof is similar to that of Theorem 8.2.9 and hence
omitted. This completes the proof.
Consider the case when q(t) behaves like a positive constant multiple
of to:, a -.\ + 1 > O. Lemma 8.2.7 shows that for every positive increasing
solution x(t) of equation (8.2.2), x(t) ~ mt(<>+2)/(A+1), t ~ to for some
constant m > O. On the other hand, it is easily seen that the equation
y"(t) = ct<>y-A(t), t ~ to where c > 0 is a constant, admits a positive
increasing solution y(t) given by
y(t)
( .\
1)2
] l/(A+1)
(a+2)(a-.\+1)
t(<>+2)/(A+l)
for
> to.
-
ct a ,
Clearly, this y(t) and p(t) =

t ~ to satisfy (8.2.87). Therefore,
Theorems 8.2.15 and 8.2.16 give the next result, which can be regarded as
an analogue to Corollary 8.2.3.
Corollary 8.2.5. Let a> .\-1. Suppose q(t) = eta [l+E(t)J, where e is
a positive constant, E(t) E C([to, (0), lR), limt--+oo E(t) = O. If 0 < .\ < 1,
then
l/(A+l)
x(t) rv
e(.\ + 1)
t(a+2)/(A+1) as t --+ 00.
2]
[(a+2)(a-.\+1)
Finally, we present the following result.
Theorem 8.2.17. Let x(t) and y(t) be positive increasing solutions

of equations (8.2,2) and (8.2.50), respectively, satisfying (8.2.85). Suppose
conditions (8.2.79), (8.2.86), (8.2.87),
/00 (100 y-2(U)dU)
-1
y-2(s) Ip(s)y3-A(S)
(f.oo y-2(U)dU)
2 _
RI ds <
00
Chapter 8
614
and either (8.2.89) or (8.2.90) hold. Then, x(t)
y(t) as t --+ 00.
Proof. The proof is similar to that of Theorem 8.2.16.
8.2.3. Positive Solutions of Singular EmdenFowler-Type Systems

Here we shall discuss positive solutions of the singular Emden-Fowlertype system (8.2.3), where
(i)
A > 0, 11
> 0 are constants,
(ii)
p(t), q(t) E C([to, 00), JR),
(iii) p, q have unbounded support,
(iv) the improper integrals P(t) = f tOC p(s)ds and Q(t) =

converge for t 2: to, P(t) 2: 0 and Q(t) 2: 0 for t 2: to, and
(v)
PQ has unbounded support.
hoc
q(s)ds
A vector function (x,y) E C 2([to,00),JR) x C 2([to,00),JR) is called a

positive solution of the system (8.2.3) when it satisfies (8.2.3) and x(t) >
0, y(t) > 0 for t 2: to. In what follows, we shall provide sufficient
conditions which ensure the existence of positive solutions (x, y) of the
system (8.2.3) satisfyin~
lim x(t) = 0,
t-'too
lim y(t) = 0,
t-'too
and
{
lim x(t)
= ,
lim y(t)
= 0,
t--l-oo
t-'toc
=0
lim x'(t)
t-'toc
lim y' ( t) = 0
(8.2.101)
t--l-OC
lim x'(t) = 0
t-'too
lim y' (t)
t-'too
= 0,
(8.2.102)
where f > 0 is a constant. For this, we need the following basic lemma.
1
1
1
P(s)ds < 00,
(8.2.103)
Q(s)ds < 00
(8.2.104)
sP(s)Q(s)ds < 00.
(8.2.105)
to
00
to
and
00
to

Then for any constants f > 0, m
solution (x, y) satisfying
615
> 0 the system (8.2.3) admits a positive
= = f,
lim y(t) = m,
t ..... =
lim x(t)
t .....
lim x'(t) = 0
t ..... =
lim y'(t)
t ..... oo
(8.2.106)
=0
and x'(t) < 0 and y'(t) < 0 for t::::: to

Proof. It is easy to verify from condition (iv) that (x, y) is a positive
solution of (8.2.3) satisfying (8.2.106) if and only if it solves the system
x(t) = f
y(t) = m
100 P(s)y-:>"(s)ds - >.100 (1= P(U)y-:>..-l(U)y'(U)dU) ds

(8.2.107)
00
00
+ 1= Q(s)x-I'(s)ds - JL 1 (1 Q(u)x-l'-l(u)x'(U)dU) ds
.
(8.2.108)
for t::::: to. First, we solve this system in some neighborhood of infinity,
say, t::::: T ::::: to. Choose positive constants c, k and T::::: to so that
>'m-:>..-l (rl' +k
00 P(S)dS) ::; c,
JLf-l'-l (m-:>.. +c
00 Q(S)dS) ::; k,
00 P(s)ds+>'m-:>..-l (rl' + k 00 P(S)dS) (00 sP(S)Q(S)dS) ::; f,

f-I' 00 Q(s)ds+Mf-l'-l ( m-:>" + c 00 Q(S)dS) (00 SP(S)Q(S)dS) ::; m.
m-:>"
Consider the set X of all functions (x, y) E C 1 ([T, 00), ]R+) X C 1 ([T, 00 ),
]R+) = Y, where components satisfy the inequalities f::; x(t) ::; 2f,
m ::; y(t) ::; 2m,
o< -
x'(t) ::; m->' P(t)
o ::; - y'(t)
::; rI'Q(t)
100
+ 100
+c
P(s)Q(s)ds
P(s)Q(s)ds
for t::::: T. Clearly, X is a nonempty closed convex subset of the Fnkhet

space Y. Define the mapping F: X -+ Y by F(x,y) = (Xl,Yl), where
Xl(t)
1= P(s)y-:>"(s)ds - >.100 (1 00 P(U)y-:>..-l(U)y'(U)dU) ds

-P(t)y-:>"(t) + >.100 P(s)y->.-l(s)y'(s)ds
=f+
x~(t) =
616
Chapter 8
and
Yl(t)
m+
00
1 (1
fJ
Q(s)x-IL(s)ds -
00
00
Q(u)X- IL - 1 (u)X'(U)dU) ds
y~(t) = -Q(t)x-~'(t) +fJ.lX) Q(S)X- IL - 1 (S)X'(s)ds

for t 2:: T. We shall show that F maps X continuously into a relatively
compact set of itself.
To prove F X
00
X, let (x, y) E X. Since
P(u)y->--l(u)y'(u)du
:::; m->--l
[rlL 1 P(u)Q(u)du + k 1 P(u) (1

00
:::; m->--l [C-IL
00
00
P(u)Q(u)du + k
for s 2:: T, we obtain
IxW)1 :::; m->- P(t)

:::; m->- P(t)
+ Am->--l
+c
00
(1
00
00
P(~)Q(~)d~) dU]
P(U)dU)
(LX) P(U)Q(U)dU)]
(rlL + k l"" P(S)dS) (1
P(s)Q(s)ds
for
00
P(S)Q(S)dS)
2:: T
and
o :::; x(t) -
:::; m->- loo P(s)ds + A
1 (1
00
:::; m->- loo P(s)ds + Am->--l

+k
1 (1
00
00
P(U)dU)
:::; m->-looP(s)ds
:::; C for
(1
00
P(U)y->--i(u)IY'(u)ldU) ds
[rlL 1 (1
00
00
00
P(U)Q(U)dU) ds
P(u)Q(U)dU)dS]
+ Am->--I (C-IL+ k looP(s)ds ) (lCOsP(s)Q(s)ds )
t 2:: T.
The estimates for YI (t) and y~ (t) can be obtained similarly. Thus, we find
that F X c X. Furthermore, it is easy to show (see previous arguments)
that the mapping F is continuous and F X is compact. Therefore, the
Schauder-Tychonov theorem implies that F has a fixed element (x,y) E
X. Hence, the system (8.2.107), (8.2.108) admits a positive solution (x, y)
in [T,oo).
617
Note that the derivative (x', y') is given by
x'(t)
y'(t)
1=
- P(t)y->'(t)
+A
Q(t)x-Il-(t)
+ /-t
00
P(s)y->'-l(s)y'(s)ds
(8.2.109)
Q(s):r;-Il--l(s)x'(s)ds
(8.2.110)
for t ~ T. Now, we claim that x'(t) < 0, y'(t) < 0 in [T,oo). Suppose
not, then X'(7) = 0 for some 7 ~ T. It follows from (8.2.109) that
A
P(s)y->.-l(s)y'(s)ds = 0, i.e., P(t)y'(t) == 0 in [7,00). Multiplying
(8.2.110) by P(t), we find that P(t)Q(t) == 0 in [7,00). This contradicts
condition (v). Hence, x'(t) < 0 in [T,oo). Similarly, we can show that
y'(t) < 0 in [T,oo).
I:"
Next, we extend (x, y) to the left as a solution of the system (8.2.3).

Let I c [to, 00) be the maximal interval of existence of (x, y). It is
clear that (8.2.107) - (8.2.110) are still valid for tEl. We claim again
that x'(t) < 0, y'(t) < 0, tEl. In fact, if this is not true, we can find
7 E I, 7 < T such that x'(t) < 0, y'(t) < 0 for t > 7 and either x'(t) or
y'(t) vanishes at 7. Suppose X'(7) = 0, then putting t = 7 in (8.2.109),
we find P(t)y'(t) == 0 in [7,00), and therefore, multiplying (8.2.110) by
P(t) we arrive a.t a contradiction as before. Hence, x'(t) < 0, y'(t) < 0
for tEl.
From the above observation it is clear that I coincides with the whole
interval [to,oo), and therefore x'(t) < 0, y'(t) < 0 for tEl = [to,oo).
Now, for the case AJL < 1 we shall prove the following results which
guarantee the existence of positive solutions of (8.2.3) satisfying (8.2.101)
or (8.2.102) with f> O.
Theorem 8.2.18. Let A/-t < 1 and conditions (8.2.103) and (8.2.104)
hold. Further, suppose that
(8.2.111)
and
(8.2.112)
Then the system (8.2.3) has a positive solution (x, y) satisfying (8.2.101).
Proof.
Clearly, the assumptions of Lemma 8.2.9 are fulfilled. Thus,
for n E IN there exists a positive solution (xn,Yn) E C 2([to,00),1R+) x
C2([to, 00), 1R+) = Y of the system (8.2.3) satisfying
x~(t)
< 0,
y~(t)
< 0 for t 2. to
(8.2.113)
618
Chapter 8
and limHCXl xn(t) = lin, limHCXl Yn(t)

limHCXl y~(t) = O. Moreover, for t:::: to,
Xn(t) =
=
Yn(t) =
l (1=
~ +l
~ 1= (1=
~ +1
~+
CXl
CXl
p(U)y;:A(U)dU) ds
P(s)y;;A(s)ds - A
= lin, limHCXl x~(t) = 0, and
1=(1= P(U)y;;A-l(U)Y~(Zl)dU)
ds
(8.2.114)
q(U).T;:I'(U)dU) ds
00
Q(s)x;;!l(s)ds - ft
1 (1=Q(U)Xr/,-l(u)x~
00
(U)dU) ds.
(8.2.115)
Hence, (8.2.113) .- (8.2.115) give
Xn(t) :::: y;;A(t)

Yn(t) :::: x;;l'(t)
1
1
00
P(s)ds,
t:::: to
(8.2.116)
00
Q(s)ds,
t:::: to.
(8.2.117)
Thus, it follows from (8.2.114) and (8.2.117) and the decreasing nature of
Yn (t) that for t:::: to,
-x~(t) =
00
p(s)y;;A(s)ds :S
<
or
_ ( x1-A!l(t))' <
1 - Aft
-
00
Integrating (8.2.118) from t to
00
Ip(s)1
x~!l(t)
ip(s)1
~
(1
1=
(100
s
00
Ip(sll
Q(U)dU) -A
(1
Q(u)du
and letting
cY0
x~l'(s)ds
Q(U)dU) -A ds,
)-A ds.
(8.2.118)
-+ 00, we get
Thus, the sequence {x n } is uniformly bounded on each compact subset of

[to, 00). Moreover, by the same computation the sequences {x~}, {Yn}
and {Y~} are a.lso uniformly bounded on each compact subset of [to,oo).
Hence, by Ascoli-Arzela's theorem we can find a subsequence {( Xni , Yni)}
of {(xn' Yn)} and a function (x, y) E Y to which {(Xni' YnJ} converges
uniformly on each compact subset of [to, 00). Inequalities (8.2.116) and

(8.2.117) show that x(t) > 0, yet) >
equations
1 (1
~i + 1= (1
~i +
Xni (t)
Yni(t) =
00
00
00
in [to, (0). Let
ni
--+
619
00
in the
2: to
p(u)y;,A(U)dU) ds,
q(u)X;;(U)dU) ds,
t 2: to
Now, the Lebesgue dominated convergence theorem asserts that (x, y) is a

positive solution of the system (8.2.3) satisfying (8.2.101). This completes
the proof.
Theorem 8.2.19. Let All < 1, (8.2.111) and ft':' slq(s)lds < 00 hold.
Then for any constant . > 0, the system (8.2.3) admits a positive solution
(x,y) satisfying (8.2.102).
Proof. As in Theorem 8.2.18, we consider the positive solutions (xn' Yn),

n E IN" of the system (8.2.3) such that x~(t) < 0, y~(t) <
for t 2:
to and limt-+ooxn(t) = ., limt-+ooYn(t) = lin, limt-+=x~(t) = 0,
limt-+oo y~(t) = 0. Note that inequality (8.2.117) remains valid. Thus, we
have
l-AI-'(t)
.l-AI-'
A
::; - - , 1 - Jl
1 - /\11
xn
100
t
slp(s)1
(100 Q(u)du)-A ds,

s
t 2: to
Now, it is easy to see that xn(t) 2: for t 2: to, and hence by the above
inequality, we get the following estimate for Yn (t)
Yn(t) ::; -1
n
+ -1-'
1=
t
slq(s)lds,
t 2: to
Finally, using similar arguments as in Theorem 8.2.18, we arrive at the

desired conclusion. This completes the proof.
When AJ.L 2: 1, it is unknown whether or not the system (8.2.3) has

such positive solutions. However, similar manipulations as in Theorem
8.2.18 give the following nonexistence criterion.
Theorem 8.2.20. Let All 2: l.
(Id If conditions (8.2.104) and (8.2.111) hold, then the system (8.2.3)
admits no positive solution (x, y) satisfying
lim x(t) = 0,
t-+oo
lim x'(t) = 0,
t-+=
lim yet) = Cl E IR o,
t-+oo
x'(t) < 0, y'(t) < 0
Cl
lim y'(t) =
t-+=
is a constant
for all large t.
(8.2.119)
Chapter 8
620
(h) If conditions (8.2.103) and (8.2.112) hold, then the system (8.2.3)
admits no positive solution (x, y) satisfying
{
lim y(t)
lim x(t)
= C2
t-+oo
t-too
x'(t) < 0,
0,
lim y'(t)
IRo,
is a constant
t.......:;,oo
y'(t) <
C2
lim x'(t)
0,
t-H)O
for all large
(8.2.120)
t.
Proof. We only consider (Id, because (12) can be treated similarly. Let
(x, y) be a positive solution of the system (8.2.3) with the required property
(8.2.119). As in Lemma 8.2.9, we find that it satisfies (8.2.107) and (8.2.108)
with C = 0 and m = y(oo) E IRo. Now choose T 2 to so large that
x'(t) < 0, y'(t) < 0 for t 2 T. Then as in Theorem 8.2.18, we get
x(t)
<
::;
1 (1
1 [1 (1
1
(1
00
p(u)y->-'(U)dU) ds
Ip(u)1
00
x-)'I'(t)
x 1 -),I'(t)
00,
00
00
i.e.,
Letting t -+
00
00
00
Q(V)dV) -), x-)'I'(U)dU] ds
(s - t)lp(s)1
slp(s)1
(1
00
00
Q(U)dU) -), ds,
Q(U)dU) -), ds,
2 T.
we get a contradiction. This completes the proof.
The following examples are illustrative.
Example 8.2.3.
q satisfying
Consider the system (8.2.3) with the functions P and
for t 2 1, where a, {3, ai, bi, i

(I)
Let AIL
= 1,2
are positive constants.
< 1. From Theorem 8.2.18, we find that if

a - A{3
> 0 and {3 -
lLa
> 0,
(8.2.122)
then the system (8.2.3) has a positive solution (x, y) satisfying (8.2.101).
In particular, the singular Emden-Fowler system
(8.2.123)
621
where D: and {3 are positive constants, has a positive solution (x, y) of

the form
x(t) = ar'Y'
{
(8.2.124)
yet) = br'Y2,
where a, b, "Ii, i = 1,2 are positive constants, if and only if condition
(8.2.122) holds. In fact, (x, y) of the form (8.2.124) becomes a positive
solution of (8.2.123) if and only if the system
+ 1)
b"f2b2 + 1)
a"llb1
b->',
- "11 - 2
>""12 - 2 - 0:
a- IL ,
"12 - 2
P,'Y1 - 2 - {3
has a solution (a, b, "11, "(2).

(II) Let >..p, > 1. It is easy to see that if
0: - >"{3 > 0
or
{3 - p,o: > 0,
(8.2.125)
then the system (8.2.123) never has a positive solution (x, y) of the form
(8.2.124). On the other hand, Theorem 8.2.20 concludes that if condition
(8.2.125) holds, then the system (8.2.3) with (8.2.121) does not have any
positive solution (x, y) satisfying (8.2.101).
Example 8.2.4.
q satisfying
Consider the system (8.2.3) with the functions p and
for t 2 1, where ai, bi, i = 1,2 are positive constants, 0:1 > 0:2 > 0 and
{31 > (32 > 0 are parameters. Let >..p, < 1. Then it follows from Theorem
8.2.18 that there exist O:i = D:i(>",p,) > 0 and (3i = (3i(>",p,) such that
the system (8.2.3), with this D:i(>", p,) and (3i(>", p,) has a positive solution
(x, y) satisfying (8.2.101). To see this it suffices to notice the fact that for
any given >.., p, satisfying >..p, < 1 inequalities
> 0,
{32 - P,0:1 > 0,
0:2 - >"{31
> 0
{31 - {32 > 0
0:1 - 0:2
admit a positive solution O:i, {3i, i = 1,2. Here, we may adopt the following
known results from convex analysis. For each n x n matrix M exactly
one of the following two cases holds:
(M 1 ) There exists n-vector satisfying
M > 0 and > o.
(M 2 ) There exists n-vector ", -=1= 0 satisfying M", < 0 and ",20.
(The order relation v 2 W [v > w] for vectors v = (Vi), W = (Wi)
defined as Vi 2 Wi [Vi> Wi] for all i).
IS
622
Chapter 8
Consider the system (8.2.3) with
Example 8.2.5.
(t)
for t 2' 1, where
p(t)
(1
E,
+ sin t )
t 2+ < '
q( t) =
is are positive constants.

0 (C 2 -
q(t) = 0 (C 2 -
E)
1=
1=
ii ) ,
_ (1 + sin
t2+ii
Let AfJ
t)
< 1. Then, we find
P(s)ds 2'
CIt
Q(s)ds 2'
C2 C1 - ii
1-<
as t --+ 00 for some positive constants C1, C2. Therefore, Theorems 8.2.18
and 8.2.19 assert that system (8.2.3) has a positive solution (x, y) satisfying
(8.2.101) if A(1 + 5) < E and /1(1 + E) < 5 and that the system (8.2.3)
has a positive solution (x, y) satisfying (8.2.102), > 0 if A(1 + 5) < c.

1. The results of Section 8.1 except Theorem 8.1.14 and Lemma 8.1.1
are taken from Naito [6-8], while Theorem 8.1.14 and Lemma 8.1.1 are duE'
to Phil os et. a1. [9J.
2. It is known [3, theorem IJ that there is a class X of f(:r) which
contains the function f(:c) = Ixl"Y sgn x h 2' 1), such that if equation
(8.1.1) with f E X has a nonoscillatory solution, then either (8.1.66)
holds, or liminfHex,(l/t) It~
q(u)duds = -00. Thus, for the special
case when f(x) belongs to X and q(t) satisfies
It:
I1 l
liminf t .....
to
to
q(u)duds > -
(8.3.1)
00,
if equation (8.1.1) has a llonoscillatory solution, then (8.1.66) holds. In

Theorem 8.1.13, however, no condition on f is needed except that
xf(x) > 0 and
f'(x) 2' 0
for
i= 0
(8.3.2)
and (8.1.65) is weaker than (8.3.1).

3. It is also known [3, theorem 2J that there is a class Y of f(x)
which contains the function f(x) = Ixl"Y sgn x (0 < I < 1), such that
if equation (8.1.1) with fEY has a nonoscillatory solution, then either
(8.1.66) holds, or limHCXl(1/t)
q(u)duds = -00. This means that
when f(x) belongs to Y and q(t) satisfies (8.1.65), if equation (8.1.1)
has a nonoscillatory solution, then (8.1.66) holds. Note that no condition
It: It:
Asymptotic behavior of solutions of certain differential equ8,tions
623
on f is needed in Theorem 8.1.13 except (8.3.2). However, instead of the

asymptotic condition as t -+ 00, the nonoscillatory solution is required.
4. The results of Section 8.1 can be extended to equations of the form
(a(t)x'(t))'
+ q(t)f(x(t))
0,
where a(t) E C([to, (0), JR+). Also, to equations with deviating arguments
of the form
xl/(t) + q(t)f(x[g(t)]) = 0,
where g(t) E C([to,oo),JR) and
to the reader.
limHoog(t) =
00.
The details are left
5. It would be interesting to obtain results similar to those presented in

Section 8.1 for equations of the form
(lx'(t)la-1x'(t))' + q(t)lx(t)I~-lX(t) = 0,
where 0: > 0, (3 >

are constants and q(t) E C([to, (0), JR), and also for
the more general equations
(lx'(t)la-lX'(t))' + q(t)f(x(t)) =
and
(lx'(t)la-1x'(t))'
+ q(t)f(x[g(t)]) =
0,
where g(t), q(t) E C([to, (0), JR), f E C(JR, JR) and limHOO g(t) =
00.
6. The results on the positive solutions of equation (8.2.1) are taken

from Tallaferro [10], while the results on the positive decaying solutions and
asymptotic behavior of solutions of singular Emden-Fowler equation (8.2.2)
are due to Usami [12,14]. The results concerning the positive solutions
of singular Edmen-Fowler-type system (8.2.3) are from Usami [13]. For
several other related works, see Tallaferro [11].
7. The results of Section 8.2 can be extended rather easily to more
general equations of the form
(a(t)x'(t))'
+ i5q(t)x- A (t) =
and equations with deviating arguments of the type
(a(t)x'(t))' + i5q(t)x- A [g(t)]

where
a(t)
>.
0,
and q are as in equations (8.2.1) and (8.2.2), <5 = 1 and
C([to, (0), JR+).
8. It is desirable to extend some of the results of Section 8.2 to equations

of the form
624
Chapter 8
where 0 = 1, a and), are positive constants, and q(t) E C([to, (0), lR).
8.4. References
1. G.J. Butler, On the oscillatory behavior of a second order nonlinear differential equation, AIlIl. AJat. Pura Appl. 105(1975), 73-92.
2. G.J. Butler, Oscillation theorems for a nonlinear analogue of Hill's equation, Quart. J Math. 27(1976), 159-171.
3. G.J. Butler, Integral averages and the oscillation of second order ordinary
differential equations, SIAM J. Math. Anal. 11(1980), 190--200.
4. T. K ura, Oscillation theorems for a second order sublinear ordinary differential equation, Proc. ArneI'. Math. Soc. 84(1982), 535-538.
5. M.K. Kwong and J.S.W. Wong, On the oscillation and nonoscillation
of second order sublinear equations, Proc. Amer. IvIath. Soc. 85(1982),
547-551.
6. M. Naito, Asymptotic behavior of solutions of second order differential equations with integrable coefficients, Trans. Amer. Math. Soc.
282(1984), 577-588.
7. M. Naito, Nonoscillatory solutions of second order differential equations
with integrable coefficients, Pmc. Amer. Math. Soc. 109(1990), 769-774.
8. M. Naito, Integral averages and the asymptotic behavior of solutions
of second order ordinary differential equations, J. Math. Anal. Appl.
164(1992), 37{}-380.
9. Ch.G. Philos and I.K. Purnaras, Asymptotic behavior of solutions of
second order nonlinear ordinary differential equations, Nonlinear Analysis
24(1995), 81-90.
10. S.D. Taliaferro, On the positive solutions of y" +<f;(t)y->' = 0, Nonlinear
Analysis 2(1978), 437-446.
11. S.D. Taliaferro, Asymptotic behavior of solutions of y" = <f;(t)f(y),

SIAM J. Math Anal. 17(1981), 853--865.
12. H. Usami, On positive decaying solutions of singular Emden-Fowler-type
equations, Nonlinear Analysis 18(1991), 795-803.
13. H. Usami, Positive solutions of singular Emden-Fowler-type systems, Hiroshima Math. J. 22(1992), 421-431.
14. H. Usami, Asymptotic behavior of positive solutions of singular EmdenFowler-type equations, J. Math. Soc. Japan 40(1994), 195-21l.
15. D. Willet, On the oscillatory behavior of the solutions of second order
linear differential equations, Ann. Polon. Math. 21(1969),175-194.
Chapter 9
Miscellaneous Topics
9.0. Introduction
This chapter contains some special results. In Section 9.1, we shall
extend the Sturm-Picone theorem, obtain nonoscillation theorems for perturbed second order nonlinear differential equations, and present a nonlinear Picone type identity to enable us to prove some Sturm-Picone type
comparison theorems for nonlinear equations. Section 9.2 is devoted to the
study of nonoscillatory solutions of forced differential equations of second
order. In Section 9.3 we shall present some limit cycle criteria and discuss
its related properties for nonlinear second order differential equations. Finally, in Section 9.4 we shall present some properties of solutions of very
general second order differential equations.
9.1. Comparison and Nonoscillation Results

The classical Sturm-Picone theorem compares equations
(a(t)x'(t))'
and
+ q(t)x(t) =
(9.1.1)
(al(t)y'(t))' + ql(t)y(t) = 0,
(9.1.2)
where a(t), al(t) E CI([to, (0), JR+), q(t), ql(t) E C([to, (0), JR). It states
that if
a(t)
:s:
al(t)
and
q(t)
ql(t)
on
[tl, t2J C [to, (0),
to> 0
then if any solution y(t) of (9.1.2) has two (or more) zeros on [tl, t2],
every solution of (9.1.1) also has at least one zero on [t l , t2J.
We shall compare the oscillatory behavior of solutions of (9.1.1) with
the solutions of the equation
(a(t)p(t)y'(t))'
+ q(t)p(t)y(t)
0,
(9.1.3)
626
Chapter 9
where p(t) E Cl([to, oo),lR). For this, we shall need the following lemma.
Lemma 9.1.1. If there exists a function y(t) E C l ([t l , t2J, lR) with y
not identically zero, and y(tl) = y(t2) = 0 such that
t2
[a(s)(y'(s))2 - q(S)y2(S)] ds :s; 0,
t,
then every solution of equation (9.1.1) except a constant multiple of y(t)

vanishes at some point of (t l , t2).
In our first result we shall prove a Sturm-type theorem concerning the
existence of zeros on the interval ttl, t2J.
Theorem 9.1.1. Suppose there exists a solution y(t) of equation (9.1.3)
which has no zeros on ttl, t2]' If a(t)p(t) 2: 0 and (a(t)p'(t))':S; 0 on
[t 1, t2J, then each solution x(t) of equation (9.1.1) can have at most one
zero on [t1, t2].
Proof. Suppose there exists a solution x(t) of equation (9.1.1) which
has zeros at t = 71 and t = 72 on the interval [tb t2]. Assume without
loss of generality that x(t) > 0 on (71,72). Then,
0=
1T2 (a(s)p(s)x(s)x'(s))'ds
T2 [12"a( s )p' (s) dsd (x 2(s)) +a(s )p( s )(x' (s))2 +p(s )x(s )(a(s )x' (s))'Jds
1
T1
T1
= -
~ 1T2 (a(s )p'(s))' x 2( s )ds+ 1T2 [a(s )p(s )(x' (s) f

'1
-q(s )p(s )x2(s)] ds.
'1
Thus, we have
1~2
[a(s)p(s)(x'(s))2 - q(s)p(s)x2(s)] ds :s; O.
(9.1.4)
Hence there exists a function x(t) E C 1([7l,72J,lR) vanishing at t = 71

and t = 72 and satisfying inequality (9.1.4). An application of Lemma
9.1.1 shows that every solution y(t) of equation (9.1.3) has a zero on
[71,72]. This contradiction proves the theorem.
Remark 9.1.1. An obvious consequence of Theorem 9.1.1 states that if

(9.1.3) is nonoscillatory, then so is equation (9.1.1).
Theorem 9.1.2. Subject to the same hypotheses of Theorem 9.1.1 any
solution of (9.1.3) has a zero on any interval containing two zeros of any
Miscellaneous topics
627
Proof. Suppose to the contrary that a solution of equation (9.l.3) has no

zeros on such an interval. Then an application of Theorem 9.l.1 yields a
contradiction.
Example 9.1.1.
((sint)y'(t))'
+ (k 2 sint)y(t) =
0,
k > 2.
(9 .l.5)
We shall show that any solution of (9.l.5) has a 2em on the interval [0,7f].
For this, we compare (9.l.5) with the equation x"(t) + k 2:r(t) = 0, which
has solutions with two zeros on the interval [0 + E, 7f - E] for a sufficiently
small E > O. Here a(t) = 1, q(t) = k 2 and p(t) = sint. Thus,
a(t)p(t) = sint > 0 and (a(t)p'(t))' = -sint < 0 on [E,7f - E]. Now by
Theorem 9.l.2 every solution of (9.l.5) has a zero on [E,7f - Elo and hence
on [0,7f].
Next, we shall consider the equation
(a(t)x'(t))'
+ q(t)x(t) =
g(t)f(x(t)),
(9.l.6)
where the functions a(t) and q(t) are as in (9.l.1), g(t) E l[to,OO)
locally, f(x) E C(JR, JR) and xf(x) > 0 for x of O.
Theorem 9.1.3. Suppose x(t) is a solution of (9.l.6) such that X(tl) =
0= X(t2)' t2 > tl > to. If there exists a function p(t) E C 1([tl, t2], JR)
such that a(t)p(t):::: 0, (a(t)p'(t))' ::; 0, g(t)p(t) ::::
on [tl' t2] and
g(t)p(t) E l[to,oo) locally, and if a1(t)::; a(t)p(t) and q1(t):::: q(t)p(t)
on [t1' t2], then allY solution of equation (9.1.2) will have a zero on [t1, t2]'
Proof. Proceeding as in Theorem 9.1.1, we have
t2 (a(s)p(s)x(s)x'(s))'ds
itl
11t2 [(a(s)p'(s))'x2(s) - 2g(s)p(s)x(s)f(x(s))] ds
-'2
t,
t2 [a(s)p(s)(x'(s))2 _ q(s)p(S)X2(S)] ds.
it l
t2 [a(s)p(s)(x'(s))2 _ q(s)p(s)x2(s)] ds
itl
::; O.
The rest of the proof is similar to that of Theorem 9.l.1 and hence omitted.
The following corollary is immediate:
Chapter 9
628
Corollary 9.1.1.
Subject to the hypotheses of Theorem 9.1.3 if any
solution of (9.1.2) has no zero on the interval [t I ,t 2 ], then any solution of
equation (9.1.6) will have at most one zero on [tI' t2]'
Example 9.1.2. Consider the differential equations
y"(t)
and
(2x'(t))'
+ y(t) =
+ x(t) =
(9.1.7)
(sin t)x 3 (t).
(9.1.8)
Here a(t) = 2, q(t) = 1, g(t) = sint and f(x) = x3 . We claim that any
solution of (9.1.8) has at most one zero on the interval ['if /3, 7T /2]. For this,
we choose p(t) = sint. Then, a(t)p(t) = 2sint ~ 1 = aI(t), q(t)p(t) =
sint S 1 = ql(t), (a(t)p'(t))' = -2sint < 0 and g(t)p(t) = sin 2 t > O.
Hence, Corollary 9.1.1 is applicable, since there exists a solution of (9.1.7)
having no zero on [7T/3,7T/2).
Remark 9.1.2. If the condition xf(x) > 0 for x f. 0 in Theorem 9.1.3
is replaced by xf(x) < 0 for x =f 0, then Theorem 9.1.3 remains true
provided g(t)p(t) ~ 0 is replaced by g(t)p(t) S O.
Now consider the differential equations
(aI(t)x'(t))'
and
+ QI(t,X(t),x'(t)) = P(t,x(t),x'(t))
(9.1.9)
(a2(t)y'(t))' + Q2(t, y(t), y'(t)) = 0,
where ai(t) E C([to, 00), 1R+), P, Qi E C([to, 00) x 1R2, 1R), i

assume that for t ~ to, (t,x,x'), (t,y,y') E [to, 00) x 1R2,
(9.1.10)
1,2. We
(9.1.11)
x
and
P(t, X, x') > O.
(9.1.12)
(9.1.13)
The following result extends the Sturm-Picone theorem.

Theorem 9.1.4. Let conditions (9.1.11) - (9.1.13) hold. If x(t) is a
solution of (9.1.9) with x(td = 0 = x(t 2) and x(t) > 0 on (tI' t 2) C
(to,oo), then every solution of equation (9.1.10) has a zero on [tI, t2)'
Proof. Let x(t) be such a solution of (9.1.9) and assume that y(t) is a
solution of (9.1.10) such that y(t) 0 on [tI, t2)' From the well-known
629
Picone identity (see Section 1.1), we have
[:~~~ (al(t)X'(t)y(t) -
a2(t)Y'(t)X(t))]'
[Q2(t'~W),y'(t))
QI(t,:~tl)'X'(t))] + [al(t)
- a2(t)](X'(t))2
+ a2(t) [ytt) (X'(t)y(t) - X(t)y'(t))] 2 + X(t)P(t, X(t), X'(t)).

An integration of this identity from iI to t2 yields the desired contradiction.

Corollary 9.1.2. Suppose conditions (9.1.11) - (9.1.13) hold.
(i) If (9.1.9) has an oscillatory solution, then every solution of (9.1.10) is
oscillatory.
(ii) If there is a solution of (9.1.10) with no zeros on [tl, t2], then no

solution of (9.1.9) which is nonnegative on [tl, t2] can vanish more than
once there.
Remark 9.1.3. \Ve can obtain analogous results when
P(t,x,x') :S 0 for
(t,x,x')
[to, (0) x IR?
(9.1.14)
In this case, however, we would need to take a solution x(t) of (9.1.9)

satisfying x(td = 0 = X(t2) and x(t) < 0 for tl < t < t2' Then every
solution of (9.1.10) must vanish on [tl, t2]'
Next, we state the following result.
Theorem 9.1.5. If in addition to conditions (9.1.11) - (9.1.13),
QI(t,X,X'):S 0 if x:SO
(9.1.15)
and equation (9.1.10) has a nonoscillatory solution, then every solution of

Remark 9.1.4. If in Theorem 9.1.5 we replace (9.1.13) by (9.1.14), then
the conclusion holds provided we replace (9.1.15) by
QI(t,x,X') 2: 0 if x2:0.
(9.1.16)
y"(t)
+ q(t)y(t)
0,
(9.1.17)
Chapter 9
630
where q(t) E C([to, 00), lR o) is such that it is nonoscillatory. By applying

Theorem 9.l.5, we find that the equation
"( )
t
x(t)
+ q(t) x 2()
t +1
is also nonoscillatory.
Now we shall consider the special case of equations (9.1.9) and (9.1.10),
namely,
(al(t)x'(t))'
+ ql(t)J'I(X(t))
and
(a2(t)y'(t))'
P(t,x(t),x'(t))
+ q2(t)12(y(t)) =
0,
(9.1.18)
(9.1.19)
where aI, a2 and P are as in equations (9.l.9) and (9.1.10), ql(t), q2(t) E
C([to, 00), lR) and J'I, 12 E C(lR, JR).
Theorem 9.1.6. Assume that condition (9.l.13) holds,
(9.l.20)
J'I(x):::;'O
if
x:::;'O
and
h(O)=O,
f~(y) >
12(y)ejO if vejO, (9.1.21)
(9.l.22)
and there is a positive constant k snch that

(9.l.23)
and
(9.1.24)
If x(t) is a solution of (9.l.18) with X(tl) = 0 = X(t2) and x(t) > 0 for
h < t < t2, then every solution of (9.l.19) has a zero on [tl, t2]'
Proof. Let x(t) be a solution of (9.1.18) with the above properties and
suppose that y(t) is a solution of (9.1.19) with y(t) ej on [tIl t2]' Since
( X ((al(t)x
t ) ) , (t)fz(y(t)) ~ a2(t)y , (t)h(x(t)))
[h
fz(y(t))
= (al(t)x'(t))'J'I(x(t))
J'
+ al(t)f{(x(t))(x'(t))2 ~
(a2(t)y'(t))'j~i~g;;
~ 2a2(t) j~i~i!;; f~(x(t))x'(t)y'(t) + a2(t) j;i~~!;; f~(v(t))(y'(t))2
631
= [q2(t) - ql(t)li?(x(t)) + h(l:(t))P(t,X(t), X'(t)) + a2(t)f~(y(t))
[( h(X(t)) y'(t))2 _ 2h(x(t)) f{(x(t)) x'(t)y'(t) + (f{(x(t)) x'(t)) 2]

h(y(t))
f2(y(t)) f~(y(t))
.f~(y(t))
-- a2(t)
(jii~i!W (:r'(t))2 + al(t)f{ (X(t))(.Y'(t))2
= [q2(t) - ql(t)lJl(x(t)) + h(x(t))P(t, X(t), l:'(t))

,
[h(X(t)) ,
f{(x(t)) ,
+ a2(t)h(y(t)) h(y(t)) y (t) - f~(y(t)) x (t)
+ f{(X(t))(X'(t))2 [a1(t) -
j2
a2(t)ji~~~!~n
an integration of the above identity from

contradiction.
tl
to
t2
gives the desired
Next, we state the following nonoscillation theorem.

Theorem 9.1.7. In addition to conditions (9.1.13) and (9.1.20) - (9.1.24)
assume that
(9.1.25)
and
h (x)
::; 0
if
x::; O.
(9.1.26)
If (9.1.19) has a llonoscillatory solutioll, then (9.1.18) is nonoscillatory.
Of particular interest in Theorems 9.1.6 and 9.1.7 is the fact that if

then we may have a2(t) > al(t), contrary to what is needed
in the Sturm-Picone theorem for linear equations. The following example
illustrates how Theorem 9.1. 7 can be applied in practice.
k < I,
Example 9.1.4.
Consider the differential equations
((1 + t)x'(t))' + q(t) tanh(x(t)/2) = 0,
t?: 2
(9.1.27)
and
((2t + sin t)y'(t))' + q(t)y(t)
0,
t?: 2
(9.1.28)
where q(t) E C([to, (0), lRo ), and assume that (9.1.28) is nonoscillatory.
Since f{(x) = (1/2) sech2(x/2)::; 1/2 and al(t) = 1+t?: (2t+sint)/2 =
a2(t)/2 for t?: t2, in view of Theorem 9.1.7 equation (9.1.27) is nonoscillatory.
In the following result, we shall replace condition (9.1.20) by
(9.1.29)
Chapter 9
632
Theorem 9.1.8. Suppose conditions (9.1.11), (9.1.13), (9.1.21) - (9.1.23)

and (9.1.29) hold. If x(t) is a solution of (9.1.18) with x(td = 0 = X(t2)
and x(t) > 0 on (tl' t2), then every solution of (9.1.19) has a zero
on [t l , t2]. If in addition (9.1.25) and (9.1.26) hold and (9.1.19) has a
nonoscillatory solution, then all solutions of (9.1.18) are nonoscillatory.
Proof. Proceeding as in Theorem 9.1.6, we have
)]'
[ ft(x(t))
f2(y(t)) ( k 2
al(t)x' (t)f2(y(t)) - ka2(t)y'
(t)h(x(t))
= _k2ql (t) /f(x(t) +k2 h (x(t) )P( t, x( t), x' (t)) +k2al (t)f{ (x( t) )(x' (t) f
+ kq2(t)il(x(t)) - 2ka2(t)y'(t) h
(xX~~~~~~(t)) x'(t)
+ ka (t)(y'(t))d?(x(t))fHy(t))
2
Ji(y(t))
;::: kfi (x( t)) [q2 (t) - kql (t)] + k 2h (x(t) )P( t, x(t), x' (t))
+ k2al (t)f{ (x(t)) (x' (t))2 +a2(t)
j1 ~~~~~~ [fi(x(t)) - 2kh (x(t) )h(y(t))
+ k 2f?(y(t))(x'(t))2] - k 2a2(t)f{ (x(t))(x'(t))2

kfi (x( t) )[q2(t) - kql (t)] + k 2h (x(t) )P( t, x(t), x' (t))
x x'(t)y'(t)
=
+ k 2f{(x(t))(x'(t))2[al(t) - a2(t)]
+ a2(t)f{(x(t)) [h(x(t))y'(t) - kh(y(t))X'(t)]2.
The remainder of the proof follows as before.
Remark 9.1.5.
1. It is easy to obtain many corollaries and theorems similar to those of
Corollary 9.1.2 and Theorems 9.1.5 and 9.1.7. The formulation of such
results are left to the reader.
2. It would be interesting to extend the above results to higher order

equations, to equations with deviating arguments, and to compare two
perturbed equations with some conditions on the relative sizes of the perturbations.
9.2. Nonoscillatory Solutions of

Forced Differential Equations
Here we shall discuss the existence of nonoscillatory solutions of the
forced differential equation
(a(t)(b(t)x(t))')'
+ F(t, x(t)) + e(t) =
O.
(9.2.1)
633
In particular, we shall show that if the forcing term e( t) is small in

certain sense, then equation (9.2.1) has nonoscillatory solutions provided
the associated unforced equation
(a(t)(b(t)x(t))')'
+ F(t, x(t))
(9.2.2)
ha.s the same property.

We begin by comparing the nonoscilla.tory behavior of the equations
(a(t)(b(t)x(t))')'
+ F1(t, x(t)) + el(t) =
(9.2.3)
(a(t)(b(t)y(t))')'
+ F2(t, y(t)) + e2(t) =
0,
(9.2.4)
and
a(t), b(t) E C([to, 00), lR+), ei(t)

C([to, oo) x lR+, lR+), i = 1,2.
where
C([to,oo),lR)
Fi
and
In what follows it will be convenient to distinguish the following two

cases:
rt
R[t,T]
iT
ds
a(s)'
and
7r(t) =
to
and
IX) a~:)
and
lim R[t,T]
t-+oo
00
(9.2.5)
7r(to) < 00.
(9.2.6)
In the following lemma, we let to and T be such that T ~ to, u E

C([to,oo),lR+), wE C([T,oo),lR+), HE C([T,oo) x lRo,lRo), X, Y E
C(n, lRo), where n = {(t, s) : t ~ s ~ T}, and H is nondecreasing in
the second variable.
Lemma 9.2.1.
Suppose the functions u, w, H, X
J';Y*(s)H(s,u(s))ds < 00, and

u(t)
w(t)
ht
X(t,s)
00
Y(v,s)H(v,u(v))dvds
and Y
for
satisfy
~ T,
where Y*(s) = max{Y(t, s) : s E [T, tl}. Then the integral equation
Z(t)
has a solution
w(t)
ht
X(t,s)
00
Y(v,s)H(v,z(v))dvds
E C([to, 00), lR+) such that w(t) :S z(t) :S u(t) for t ~ T.
F1(t,x)
F2(t,x)
for
x> 0
(9.2.7)
634
Chapter 9
and F2 (t, x) is nondecreasing in x. Further, suppose that there exist two

functions T]l, T]2 E C([to,oo),lR) such that T]l(t) is oscillatory,
(a(t)(b(t)TJi(t))')'
+ ei(t)
t2":to,
0,
i=1,2
(9.2.8);
and
lim b(t)r/i(t)
1-+00
0,
i=1,2
0,
(9.2.9)
when condition (9.2.5) holds, also
. b(t)
11m - (
) T];(t)
t
1-+00 If
1,2
(9.2.10)
when condition (9.2.6) holds.

If (9.2.3) has an eventually positive solution x(t), then (9.2.4) has an
eventually positive solution y(t) such that y(t) s: x(t) for all large t.
Proof. Let x(t) be a positive solution of (9.2.3) defined on [to, (0). Set
w(t) = x(t) - T]l (t). Then for t 2": to, we find
(a(t)(b(t)w(t))')'
Fl (t, x(t)) <
o.
(9.2.11)
Thus, w(t) is eventually of constant sign. If w(t) < 0 for t 2": tl for
some tl 2": to, then 0 < x(t) < T]l(t), t 2": tl which contradicts the fact
that T]l(t) is oscillatory and hence we must have w(t) > 0 for t 2": tl.
Now we need to consider two cases:
(i) Suppose condition (9.2.5) holds. It is easy to see that a(t)(b(t)lL'(t))' >
for t 2": tl and the finite limit C2 = limHoo a(t)(b(t)w(t))' 2": 0 exists.
Integrating equation (9.2.11) from t 2": tl to u 2": t and letting u -+ 00,
we obtain
a(t)(b(t)w(t))' =
C2
00
Fl(S,X(s))ds,
2": tl.
(9.2.12)
Dividing equation (9.2.12) by a(t) and integrating from tl to t, we get
b(t)w(t)
Cl
+C2R[t,h]
l -() 1=
1
t,
ass
or
b(t)x(t) = cl+b(t)T]1(t)+c2R[t,tl]+
for t 2": tl, where
t2 2": h such that
Cl
Fl(U,X(u))duds,
l 1=
t
t,
1
-()
ass
Fl(U,X(u))duds
= b(tdw(t l ) > O. In view of
(9.2.13)
(9.2.9) there is a
(9.2.14)
635
From (9.2.13), (9.2.14) and (9.2.7), we find

1 l
b(t)x(t) :::: ?C
-
+ b(t)'T/2(t) + c2R[t, tIl +
it 1
to
1
-()
00
a s
F2(u, x(u))duds,
(9.2.15)
Now by Lemma 9.2.1 it follows that there exists a continuous
solution y( t) of the integral equation
t :::: t2'
1 +b(t)ry2(t)+c2R[t, tll+
b(t)y(t) = -Cl
it 1
00
1
-()
has
F2(u, y(u))duds, t:::: t2
(9.2.16)
satisfying
b(t)x(t) :::: b(t)y(t) ::::
"2C1 + b(t)'T/2(t) + C2R[t, tIl
for
t:::: t2' (9.2.17)
Differentiating equation (9.2.16) twice, it follows that yet) is a solution

of (9.2.4) on [tl' (0). Finally, x(t):::: yet) > 0 on [t2' (0) follows from
(9.2.14) and (9.2.17).
(ii) Suppose condition (9.2.6) holds. Then we rewrite equations (9.2.3) and
(9.2.4), respectively, as
0,
(9.2.18)
O.
(9.2.19)
and
(a(t)7r 2 (t)
Since
to
(!~~ y(t)) ') / + 7r(t)F(t, yet)) + 7r(t)e2(t)

1
-~-,--,-ds
a(s)7r 2(s)
- - - -+
7r(t) 7r(t o)
00
as
t -+
00,
the above arguments in (i) apply to equations (9.2.18) and (9.2.19), and
the desired conclusion follows immediately. This completes the proof.
Remark 9.2.1. Suppose 'T/I(t) is oscillatory, (9.2.8h and either (9.2.9)

or (9.2.10) are satisfied. Let x(t) be an eventually positive solution of
equation (9.2.3). Then it is easy to verify that there are constants ki >
0, i = 1,2 and T:::: to such that k 1 ::; b(t)x(t) ::; k2 R[t, tol for
T if
condition (9.2.5) holds, and k l 7r(t)::; b(t)x(t) ::; k2 for t:::: T if condition
(9.2.6) holds. This observation shows that:
t::::
(It) If condition (9.2.5) holds, a solution x(t) of (9.2.3) with the property
limHOOb(t)x(t)jR[t,toJ = constant> 0 (respectively, limHoob(t)x(t) =
Chapter 9
636
constant> 0) can be regarded as a 'maximal' (respectively 'minimal')

positive solution of (9.2.3).
(h) If condition (9.2.6) holds, a solution x(t) of equation (9.2.3) with

the property limt--+oo b(t)x(t) = constant> 0 (respectively limHoo b(t)
x x(t)j7f(t) = constant> 0) can be regarded as 'maximal' (respectively
'minimal') positive solution of (9.2.3).
The following result establishes the existence of minimal positive solutions of equations (9.2.3) and (9.2.4).
Theorem 9.2.2. Suppose conditions (9.2.7), (9.2.8)i - (9.2.10), i = 1,2
are satisfied.
(i) Let condition (9.2.5) hold. If equation (9.2.3) has a solution x(t) such
that limHoo b(t)x(t) = constant> 0, then (9.2.4) has a solution y(t)
such that y(t) :s; x(t) for all large t and limHoo b(t)y(t) = constant> O.
(ii) Let condition (9.2.6) hold. If equation (9.2.3) has a solution x(t) such
that limHoo b(t)x(t)j7f(t) = constant> 0, then (9.2.4) has a solution
y(t) such that y(t):s; x(t) for all large t and limHoo b(t)y(t)j7f(t) =
constant> O.
Proof. It suffices to prove the statement (i). Let x(t) be a positive
solution on [to,oo) of (9.2.3) such that limHoo b(t)x(t) = c > 0, where
c is a constant. Put w(t) = x(t) - 'T71 (t). As in the proof of Theorem
9.2.1, we see that w(t) is eventually of constant sign, say, for t 2: t1 2: to.
If w(t) < 0, t 2: tl then x(t) < 7/1(t), t 2: t1 which implies that
liminfHoob(t)'T71(t) 2: c> 0, a contradiction to condition (9.2.9). Thus,
we have w(t) > 0 for t 2: tl' Proceeding as in the proof of Theorem 9.2.1(i)
and noting that limHoo b(t)w(t) = C, C2 = limHoo a(t)(b(t)w(t))' = 0,
we conclude that there exists a continuous solution y(t) of the integral
equation
1
b(t)y(t) = -Cl
2
+ b(t)'T72(t) +
it -(-) 1
t2
1
ass
00
F2(u, y(u))duds
satisfying b(t)x(t) 2: b(t)y(t) 2: (cI/2) + b(t)'T72(t) for t 2: t2, where

Cl > 0 is a constant and t2 > tl is sufficiently large. Now it is easy to
see that y(t) is a minimal positive solution of (9.2.4). This completes the
proof.
Next, for the existence of maximal positive solutions of equations (9.2.3)

and (9.2.4), we only state the following result.
Theorem 9.2.3. Let conditions (9.2.7) and (9.2.8)i hold,
b(t)
Inll -R[j'T71(t)
t--+oo t, to
0,
b(t)
11m -R[j'T72(t)
t--+oo t, to
(9.2.20)
637
when condition (9.2.5) holds, and
(9.2.21)
when condition (9.2.6) holds.

that limHoo b(t)x(t)/ R[t, tol = constant> 0, then (9.2.4) has a solution
y(t) such that y(t) ::::: x(t) for all large t and limHoo b(t)y(t)/ R[t, tol =
constant > O.
that limHoo b(t)x(t) = constant> 0, then (9.2.4) has a solution y(t)
such that y(t) ::::: x(t) for all large t and limHoo b(t)y(t) = constant> O.
Now we shall consider equations (9.2.1) and (9.2.2) where a(t), b(t) E
F E C([to,oo) x JR+, JR+) and
C([to, 00), JR+), e(t) E C([to, 00), JR),

F(t, x) is nondecreasing in x.
First we state necessary and sufficient conditions for equation (9.2.2) so

that it possess maximal and minimal positive solutions.
Theorem 9.2.4. (i) Suppose condition (9.2.5) holds. Then, (9.2.2) has
a maximal (respectively minimal) positive solution if and only if
XJ
F (s, bts)R[s,to1 ) ds < 00
(/00 R[s,tolF (s, bts)) ds < 00)
for some constant c > O.

(ii) Suppose condition (9.2.6) holds. Then, (9.2.2) has a maximal (respectively minimal) positive solution if and only if
for some constant c> O.

In particular, for equations (9.2.1) and (9.2.2), Theorems 9.2.1 - 9.2.3
yield the following results, which roughly conclude that the nonoscillatory
character of equation (9.2.2) is not affected by adding a small forcing term
e(t).
Corollary 9.2.1. Suppose there exists a function 'r/(t) E C([t o, 00), JR)
with the properties
(a(t)(b(t)'r/(t))')'
lim b(t)'r/(t)
t400
+ e(t)
0,
t:::: to
when condition (9.2.5) holds,
(9.2.22)
(9.2.23)
Chapter 9
638
b(t)
11m -(-)TJ(t)
t-)OO
n t
..
= 0 when condItIoIl (9.2.6) holds.
(9.2.24)
If equation (9.2.2) has an eventually positive solution x(t), then (9.2.1) has
an eventually positive solution y(t) such that y(t) ::; x(t) for all large t.
Corollary 9.2.2. Suppose there exists a functionf)(t) E C([to, 00), JR)
satisfying condition (9.2.22) and either condition (9.2.23) or (9.2.24).
that limt-400 b(t)x(t) = constant> 0, then (9.2.1) has a solution y(t)
such that y(t) ::; x(t) for all large t and limHoo b(t)y(t) = constant> O.
that limHoo b(t)x(t)/n(t) = constant> 0, then (9.2.1) has a solution
y(t) such that y(t)::; x(t) for all large t and limHoo b(t)y(t)(rr(t) =
constant> O.
Corollary 9.2.3. Suppose there exists a function TJ(t) E C([to, 00), JR)
satisfying condition (9.2.22) and
lim Rb[(t) ]TJ(t)

t, to
when condition (9.2.5) holds,
t-400
lim b(t)T](t)
t~oo
= 0 when condition (9.2.6) holds.
that limHoo b(t)x(t)/ R[t, to] = constant> 0, then (9.2.1) has a solution
y(t) such that y(t)::; x(t) for all large t and lill1Hoo b(t)y(t)/ R[t, to] =
constant> O.
that limHCXl b(t)x(t) = constant> 0, then (9.2.1) has a solution y(t)
such that y(t) ::; x(t) for all large t and limHoo b(t)y(t) = constant> O.
Remark 9.2.2. In the above results we were concerned only with eventually positive solutions of the equations under consideration. It is clear that
these results have counterparts for eventually negative solutions provided
F i , F E C ( [to, 00) x JR - , JR - ), i = 1, 2.
9.3. Limit Circle Criteria and Related

Properties For Nonlinear Equations
In this section we shall present some sufficient conditions for the integrability of solutions of the forced second order nonlinear differential equation
(a(t)x'(t))'
+ q(t)f(x(t))
e(t).
(9.3.1)
639
We will say that equation (9.3.1) is of limit circle type if all its solutions
satisfy
I:)Q x(u)f(x(udu
<
(9.3.2)
00.
Otherwise equation (9.3.1) is said to be of limit point type.

Now we shall consider equation (9.3.1) where a(t), q(t) E C([to, (0),
lR+), e(t) E C([to, (0), lR), f E C(lR, lR), a'(t), q'(t) E ACtoc[to, (0),
a"(t), q"(t) E Lroc [to , (0) and xf(x) > 0 for x i= O.
As earlier for any h(t) E C([to, (0), lR) we let h+(t) = max{h(t),O}
and h-(t) = max{ -h(t), O}. Also, we define F(x) =
f(u)du. We
shall assume that there exist positive constants k andn and nonnegative
constants A, Band K such that
J;
2(n+1),
(9.3.3)
o ::; kxf(x) - n + 1 F(x) ::; BF(x)
(9.3.4)
and
(9.3.5)
= 1/(2(n + 1
To simplify the notation in what follows, \\Te let a

/3 = (2n + 1)/(2(n +
We use the transformation
1.
s =
it
a-(3(u)qC> (u)du
and
and
= x(t)
(9.3.6)
(. = d/ds)
(9.3.7)
y(s)
to
so that equation (9.3.1) becomes
ii + ap(t)y + Q(t)f(y)
E(t),
where p(t) = (a(t)q(t' /(aQ(t)qQ+l(t, Q(t) = (a(t)q(t(3-Q and E(t)

a(3-c> (t)e(t)q-2c> (t). Note also that /3 - a = 2/3 -1 = n/(n + 1).
In system form equation (9.3.7) can be written as

.
y=
z -
-p(t)y
~ (f ~ +(t)z - Q(t)f(y) + [P(t) - G-a)v'(t)1(H + E(t).

(9.3.8)
Theorem 9.3.1. In addition to conditions (9.3.3) - (9.3.5) assume that
1=
to
[(a(u)q(u']- d
a(u)q(u)
u <
00,
(9.3.9)
640
Chapter 9
rOO I [
lto
(1k - 0:) aa+l(u)qa+2(u)

[(a(u)q(u'j2 Id
u <
(a(u)q(u' ]'
aa(u)q<>+l(u) -
00
. to
le(u)1
d
U <
(a( v,)q( u))a
~..,-:.--:.--'-:-:--
00,
(9.3.10)
(9.3.11)
00
and
(9.3.12)
Then any solution x(t) of (9.3.1) satisfies (9.3.2) and
Proof.
Define V(y, z, s) = (1/2)z2
0:) p(t)Z2 + [pet) -
(~ -
V(s)
+ Q(t)F(y),
-Q(t)p(t)
(~ -
ft'; F(x(u))du < 00.
then
0:) p2(t)] (~yz) + E(t)z
(~Yf(Y) + Q(t)F(y)
(~- 0:) p(t)z2 + [pet) - (~ -
0:) p2(t)] (~yz) + E(t)z
1 ] [(a(t)q(t)'] (a!3(t))
+Q(t) [ (f3 - o:)F(y) - kyf(y)
a(t)q(t)
qa(t)
Applying conditions (9.3.3) and (9.3.4), we get
V(s)
[B - 2
(~- 0:)] ([(a(t)q(t),l-) (a!3(t) V

a(t)q(t)
k:
+E(t)z+ [p(t)-
qa(t)
(~-0:)p2(t)] (~yz).
Now from (9.3.5), we find
Iyzl
111
~ 2,y2 + 2,Z2 ~ 2,z2
+ AF(y) + K,
so
V(s) < [B - 2
-
+IE(t)1
Since pet)
(~- 0:)]
k
([(a(t)q(t),l-) (a!3(t).) V
a(t)q(t)
qa(t)
(~+ V) + \p(t) - (~ - 0:) p2(t)\ ~ [V + K + AQ~t)] .
= p(t)a!3(t)q-a(t),
let)
Ip(t) Ip'(t) -
we have
0:) p2(t) I
(~ - 0:) p2(t)q<>(t)a- i3 (t)i a!3(t)q-<>(t).
(~ -
641
Let T(S) denote the inverse function of s(t), we obtain by (9.3.10) that
converges. Similarly,
So
IE[T(~)lld~ =
It
cy,
IE(u)lq (u)a
to
-(3
It
(u)du -
to
le(u)1
(a(u)q(un du
converges by (9.3.11). Next observe that condition (9.3.7) implies that

Q(t) is bounded from below. Hence if we integrate V from So to s, use
the bound indicated above, apply Gronwall's inequality and then transform
the integrals from s back to t, we find that conditions (9.3.9) - (9.3.11)
are precisely those needed to insure that V(s) is bounded. Thus,
= (a(t)q(t(3-a F(y(s
(a(t)q(t(3-a F(x(t
~ Kl
for some constant Kl 2: 0, and now the conclusion of the theorem follows
from conditions (9.3.4) and (9.3.12).
In particular, for the Emden-Fowler equation
x"(t)
+ q(t)x'Y(t)
0,
(9.3.13)
where 'Y is an odd positive integer and q(t) is as in equation (9.3.1),

Theorem 9.3.1 yields the following result.
Corollary 9.3.1. Assume that
1I
00
to
and
1I
00
where ()
(2n
q"(U)
[q'(u)j2 Idu <
- - (}--qO ( u)
q>' (u)
q:(( u - () [q'; U121 q-n/(n+l) (u)du <

q
(u
+ 3)/(2(n + 1
and A = (4n
roo q-n/(n+l)(u)du
ito
(9.3.14)
00
+ 5)/(2(n + 1.
<
00,
then every solution x(t) of equation (9.3.13) satisfies
(9.3.15)
00,
If
(9.3.16)
ftC; x2n(u)du < 00.
Actually, the corollary stated above follows by using the transformation

(9.3.6) and the technique of the proof of Theorem 9.3.1 directly to equation
(9.3.13) (the n in (9.3.6) is chosen so that 'Y = 2n-1). Conditions (9.3.3)
642
Chapter 9
and (9.3.4) are satisfied if we take k = 2('1'1

(9.3.5) holds, (e.g., let A = nand k = 1).
+ 1)
o.
and B =
Clearly,
Next, we shall discuss some relationship between the nonlinear limit

circle property and the boundedness, oscillation and convergence to zero of
Theorem 9.3.2. Under the hypotheses of Theorem 9.3.1 all solutions
of equation (9.3.1) are bounded. If in addition F(:r;) > 0, x =I 0 and
a(t)b(t) -+ 00 as t -+ 00, then a.ll solutions of (9.3.1) converge to zero as
t -+
00.
Proof. From the proof of Theorem 9.3.1, we have (a(t)q(t))!3- a F(x(t)) s:

K 1. Condition (9.3.9) implies that a(t)b(t) is bounded below away from
zero, so F(x(t)) is bounded. The boundedness of x(t) now follows from
condition (9.3.5). If a(t)b(t) -+ 00 as t -+ 00, then the above inequality
shows that F(x(t)) -+ 0 as t -+ 00. Since F(x) > 0 for x =I 0, we find
that x(t) -+ 0 as t -+ 00.
> 0, N
Theorem 9.3.3. Suppose there exist constants M

and c > 0 such that
x 2 < Mxf(x)
1=
and
0, m
la 1/2 (t)q' (t )q-3/2 (t) I < m,
+ N,
Iq(t)
e(t) I
l=le(u)ldU
q(u)
< c,
-
to
<
a(u)[q'(u)J2 q-3(u)du <
>0
(9.3.17)
(9.3.18)
00
(9.3.19)
00.
to
If x(t) is a limit circle solution of equation (9.3.1), i.e., (9.3.2) holds, then
= a(u),
to
- () [x (u)] du <
q
(9.3.20)
00.
Proof. Multiply equation (9.3.1) by x(t)/q(t) and note that (a(t)x')'x =

(a(t)xx')' - a(t)[x'J2, and integrate by parts from t1 ~ to to t, to find
a(t),(
a(t1),()
()
() x t)x ()
t - -(-)
X t1 x t1
q t
q t1
it
t,
x(u)f(x(u))du
-it
it
a(u) '()'() ()
~() q u x u X u du
t, q
a((u)) [x'(u)J 2du =
t, q U
it
e((u)) x(u)du.
t, q
(9.3.21)
643
By the Schwarz inequality, we have
11: ;(~) q'(u)X'(U)X(U)dUI
<
Now from condition (9.3.17), we find
;~tl) [q'(t)fx 2(t)
::; m 2 M x(t)f(x(t))
+ N ;~tl) [q'(t)f,
which on integrating and then applying (9.3.2) and (9.3.19), gives
roo q~(u)
[q'(uWx2(u)du
(u)
itl
::; Kl <
00,
Kl is a constant.
Since
<
Ix (t )I e(t)
() _
qt
e(t) 2()
e(t) < ~'1' ()f( ())
2q (t ) x t + 2q ()
t _ 2 lVj X t x t +
(N+1) qe(t)t
2
( )'
the integral of the right-hand side of (9.3.21) converges in view of (9.3.18)

and (9.3.2). If x(t) is not eventually monotonic, let {tj}, limj---+= t.i =
00 be an increasing sequence of zeros of x'(t). Then from (9.3.21), we
have KIHl/2(tl) + K2 2: H(tl), where K2 is a constant, and H(t) =
fttl (a(u)(x'(u))2 /q(u))du. It follows that H(tj)::; K3 < 00 for all j,
where K3 is a constant, and so (9.3.20) holds. If x(t) is eventually
monotonic, then x(t)x'(t) ::; 0, t 2: tl for all sufficiently large tl 2:
to, since otherwise condition (9.3.2) will be violated. Using this fact in
(9.3.21) we can repeat the type of arguments used above to again obtain
that (9.3.20) holds.
It is known that equation (9.3.1) with f(x) = x and e(t) = 0 is

oscillatory if it is limit circle. But, this is not true in general for forced
equations. For example, Theorem 9.3.1 guarantees that all solutions of
(tx'(t))'
+ t 2 x(t)
t2
+t
for
2: 1
belong to 2[1,(0), but x(t) = 1/t3 is a nonoscillatory solution of

this equation. However, the following result holds for the forced equation
(9.3.1).
Theorem 9.3.4. Assume that f(x) is bounded away from zero if x
bounded away from zero,
IS
(9.3.22)
644
Chapter 9
and equation (9.3.1) is limit circle. Then every solution of equation (9.3.1)
either oscillates or converges monotonically to zero as t --+ 00.
x(t) > 0 for t:;:, tl :;:, to. Clearly, liminft._HXl x(t) = 0 for otherwise
condition (9.3.2) will be violated. If x(t) is eventually monotonic, we are
done, so we assume that x(t) is Hot eventuR.lly monotonic. If x(t) does
Proof.
not converge to zero as t -t 00, then there exists a constant Kl > 0

such that for any t2 > tl there exists t3 > t2 with X(t3):;:' K 1. Choose
t2 > t1 so that
L= (1:' atv)
dl) le(u)ldu <
and choose t3 > t2 such that x' (t3) = D,

equation (9.3.1) from t3 to t, we find
1
x'(t) = -()
a t
It
t3
~K1
and x( t3) :;:, K 1. Integrating
[e(u) ~ q(u)f(x(u))]du.
(9.3.23)
Another integration yields
x(t)
X(t3)
+ (ft __
(1 )d1L) ft[e(u)
13
~ ft
t3
au
/3
~ q(u)f(x(u))]du
(1" _(1,)dV) [e(u) ~ q(lL)f(x(u))]du.

/3
1~
If t4 > h is any zero of x'(t), then (9.3.23) shows that the first integral
above vanishes at t = t 4 , and then we have
x(t) is bounded from below by Kr/2 at every zero of x'(t) for

which contradicts liminf t -+= x(t) = D. The proof for the case
x(t) < 0, t:;:, tl is similar.
i.e.,
t :;:, t3,
Next, we state two results for the oscillation of unforced and forced
equations, respectively.
Theorem 9.3.5.
If in addition to the hypotheses of Theorem 9.3.4,
e(t) = 0 and f= ds/a(s) = 00, then all solutions of equation (9.3.1)
are oscillatory.
Theorem 9.3.6. Suppose the hypotheses of Theorem 9.3.4 hold except
possibly condition (9.3.22). If for every constant c > 0 and all large
T:;:, to,
and
645
li~~p [ [ atu) (ilL e(v)dv + c) dU]
00,
then all solutions of equation (9.3.1) are oscillatory.

The proofs of Theorems 9.3.5 and 9.3.6 are easy and hence omitted.
Next, we are concerned with the limit point criteria for second order
nonlinear equations of type (9.3.1).
To simplify the notation used in our next result, we define the functions
h, H: [to, 00) -+ 1R. by
h(t) = [(a(t)q(t))']a(t)q(t)
+ le(t)1
and H(t) =
[(a(t)q(t))']+
a(t)q(t)
le(t)1
+ (a(t)q(t))l/2
Theorem 9.3.7. Suppose conditions (9.3.17) - (9.3.19) hold, there exists

a constant Ml > 0 such that
F(x) ::::: Nhxf(x)
(9.3.24)
and either
L= !~:~
1~ le(u)1 (exp
or
(h)
1~ exp (-
L
1:'
exp ( -
L
U
h(V)dV) du =
(9.3.25)
h(V)dV) du <
H(V)dV) du =
00,
00,
00,
r= le(u)1 ( r
)
ltD (a(u)q(u))I/2
ltD H(v)dv du <
exp
(9.3.26)
00.
Then equation (9.3.1) is oflimit point type, i.e., there is a solution of (9.3.1)
which does not satisfy condition (9.3.2).
Proof. We will write equation (9.3.1) as the system
{ :; :
[-a'(t)y - q(t)f(x) + e(t)](l/a(t)).
If (II) holds, we define V(x, y, t)
= (1/2)a 2(t)y2(t) + a(t)q(t)F(x).
(9.3.27)
Then,
V'(t) = a(t)e(t)y + (a(t)q(t))'F(x) ;::: a(t)e(t)y _ [(a(t)q(t)),J- V.

a(t)q(t)
646
Chapter 9
Now, we have
so it follows that
~ - ([(a(t)q(t))']- + le(t)l) V - ~le(t)l.
V'es)
a(t)q(t)
(9.3.28)
From condition (9.3.25), we find
~ Lex> le(u)1 (exp
(1:
h(V)dV) ) du ::; K1 <
00.
Let (x(t), yet)) be a solution of system (9.3.27) such that (x(to), y(to)) =
(xo, yo) and V(xo, Yo, to) = Veto) > KI + L Then from (9.3.28), we get
(V(t)ex p
(1:
h(U)dU)), ;:: -
~le(t)lexp
(1:
h(U)dU).
Integrating this inequality from to to t, we obtain
Vet) exp
(1:
h(U)dU)
;:: Veto) - KI > L
Hence,
In view of Theorem 9.3.3 this shows that x(t) cannot be a limit circle
If (1 2 ) holds, we define
y2
a(t)-()
2q t
+ F(x).
Then, we have
V'et) > e(t) _ [a(t)q(t)]+ V;

1
q(t)Y
a(t)q(t) 1
However, since
e(t) I [
le(t)1
]
Iq(t)Y
(a(t)q(t))l/2
we find
V{(t)
+ H(t)Vi(t)
;::
(
1
VI
1)
+"2 '
le(t)1
-"2 (a(t)q(t))l/2
647
The rest of the proof is similar to the case (II) and hence omitted.
In our next limit point theorem we shall need the following lemma.
!VI > 0, N ::::: 0 and
Lemma 9.3.1.
Suppose there exist constants
> 0 such that
Tnl
x 2 :s; Mxf(x)+N,
(a(t)q(t))' I
Ia 1/2(t)q3/2(t)
:s; ml and
(9.3.29)
roo [(a(u)q(u))']2 du
lto
a(u)q3(u)
<
00.
(9.3.30)
If x(t) is a limit circle solution of equation (9.3.1), (i.e., condition (9.3.2)

holds), then
roo [(a(u)q(u))']2 x2(u)du
lto
a(u)q3(u)
<
00.
Proof. Clearly, by condition (8.3.30)
roo [(a(u)q(u))']2 x (u)du
lto
a(u)q3(u)
< miM
-
+
Theorem 9.3.8.
Bl ::::: 0 such that
kl
00
to
roo x(u)f(x(u))du
lto
[(a(u)q(u))']2d
u <
a(u)q3(u)
00.
Suppose there exist constants kl > 0, n > 0 and

1
:s; 2(n + 1) and 0 :s; (3 - o:)F(x) - kl xf(x) :s; B 1 F(x). (9.3.31)
In addition assume that conditions (9.3.5), (9.3.9) - (9.3.11), (9.3.18),

(9.3.19), (9.3.24) and (9.3.30) hold with the k in condition (9.3.10) replaced by kl' and
(9.3.32)
If
00,
(9.3.33)
then equation (9.3.1) is of limit point type.

Proof. Apply the transformation (9.3.6) to (9.3.1) to obtain equation
(9.3.7), and then write (9.3.7) in the form of system (9.3.8) with know
replaced by k 1 . As in Theorem 9.3.1, we define V(y,z,s) = (1/2)z2 +
648
Chapter 9
(a(t)q(t))i3- QF(y) and differentiate, to obtain
n) aQ(t)qQ+l(t)
[(a(t)q(t))'](v + ~)
-Ip(t) - (~ - n) p2(t)1 ~ [v + K+ A~]
kl
kl
Q(t)
V(s) ?: -
(~ -
-Q(t)
> _
-
-IE(t)1
z2
kl
[(f3 -
~Yf(Y)] [a(t)q(t))']- ai3 (t)
n)F(y) _
a(t)q(t)
kl
qQ(t)
[Bl + 2 (~n)] ([(a(t)q(t))']-)

(a i3 (t)) V
kl
a(t)q(t)
qQ(t)
(~n) p2(t)1 ~kl [V + K+ A~]

-IE(t)1 (V +~) .
kl
Q(t)
2
-Ip(t) -
Now define the functions G, g: [to, 00) -+ 1R by
G(t)
+ le(t)1
{ [Bl + 2 (~kl _ n)] [(a(t)q(t))']a(t)q(t)
(a(t)q(t))Q
'
+ Ip (t) -
( kl1
- n
qQ (t) I [
1 ] 1 } ai3 (t)
(t) af3(t) 1 + AQ(t) kl qQ(t)
and
K I'
g(t) = kl P (t) so that V(s)
1
kl -
n) p
+ G(t)V(s) ?: -g(t),
(vex (1: G[T(~)]d~))

p
qQ(t) I
le(t)1
af3(t)
(t) af3(t) + 2(a(t)q(t))Q qQ(t) ,
and hence
?: - g(t) exp
(1: G[T(~)]d~)
?: - Klg(t)
(fa
(9.3.34)
G[T(~)]d~) :S
since conditions (9.3.9) - (9.3.11) guarantee that exp
Kl < 00 for some constant Kl > O. In view of (9.3.10) and (9.3.11), we
have Kl
g[T(~)]d~ :S K2 < 00 for some constant K2 > O. Now let
x(t) be any solution of (9.3.1) such that V(y(so), z(so), so) > K2 + l.
Integrating (9.3.34) from So to s, we find
Is";
V(s) exp
(1: G[T(~)]d~)
?: V(So) - K2 > 1
and thus V(s)?: 1/ Kl for s;::: so. Now dividing this inequality by
(a(t)q(t))i3- Q and rewriting the left-hand side in terms of t, we get
a(t) (x'(t))2
2q(t)
+ (a(t)q(t))' x(t)x'(t) + [(a(t)q(t))']2 x2(t) + F(x(t))

kq2(t)
2k 2a(t)q3(t)
>
;1
(a(t)q(t))Q-i3.
(9.3.35)
649
If x(t) was a limit circle solution of equation (9.3.1), then since conditions (9.3.5) and (9.3.24) imply (9.3.29), in view of Theorem 9.3.3, we have
ftC;: (a(u)jq(u)) (x'(u))2du < 00. Also, by Lemma 8.3.1
r= [(a.(u)q(u))']2 x2(u)du <

lto
a(u)q3(u)
(X)
and by Lemma 8.3.1 and Theorem 8.3.3

ILOX)
:::;
(a(;;r~~))' x(u)x'(u)dul
[r=
lto
[(a(u)q~u))']2 x2(U)dU] 1/2 [

a(u)q (u)
r= a(u) [X'(UWdUll/2 <

lto q(u)
(X)
and by condition (9.3.24) and the assumption that x(t) was a limit circle
solution ItC;: F(x(u))du < 00. Now integrating (9.3.35) we get the desired
contradiction.
By combining Theorem 9.3.8 with Theorems 9.3.1 and 9.3.3 we can

obtain necessary and sufficient conditions for the equation
(a(t)x'(t))'
+ q(t)X 2n - 1 (t) =
e(t),
(9.3.36)
where n is a positive integer, to be of limit circle type, i.e., all solutions

2n [to, (0). For this, first note that conditions (9.3.3), (9.3.4)
belong to
and (9.3.31) are satisfied with k = kl = 2(n + 1) and B = Bl = O. In
addition, conditions (9.3.5) and (9.3.24) are automatically satisfied.
.c
Theorem 9.3.9. Suppose conditions (9.3.9), (9.3.11), (9.3.18), (9.3.19),

(9.3.30) and (9.3.32) hold, and
r=l[ (a(u)q(u))' ]'ldU <

ltD
a<>(u)q<>+l(u)
00.
Then (9.3.36) is oflimit circle type if and only if ftC;:[a(u)q(u)]-n/(n+l)du
< 00.
In particular, Theorem 9.3.9 for
x"(t) + q(t)x2n-l(t)
(9.3.37)
immediately gives the following corollary.

Corollary 9.3.2. If
r= [q'(u)]- du <
ltD
q( u)
00,
OX) [q'(u)F
~(
) du <
to
00,
650
Chapter 9
where m2 2: 0 is a constant and condition (9.3.14) holds, then equation

(9.3.37) is of limit circle type if and only if
(00 q-n/(n+l)(u)du <
Jto
Example 9.3.1.
00.
(9.3.38)
0,
(9.3.39)
Consider the equation
xl/(t)
+ t":C 2n -
1 (t)
where (J" is any constant and n is a positive integer. Now condition

(9.3.38) implies that (J"nl(n + 1) > 1, or
(J"
> 1 + (lin).
(9.3.40)
It is easy to check that all the hypotheses of Corollary 9.3.2 are satisfied,
and hence condition (9.3.40) is a necessary and sufficient condition for all
solutions of equation (9.3.39) to belong to 2n[to,00).
9.4. Properties of Certain

Consider the second order differential equation
(a(t)7/;(X(t))X'(t))'
+ q(t)f(x(t))
e(t),
(9.4.1)
where a(t) E C1([to,oo),JR+), e(t), q(t) E C([to,oo),JR), to 2: 0, 7P(X) E

C1(JR, JR), f(x) E C(JR, JR) and xf(x) > 0 for x of. O.
We shall show that under appropriate conditions on a(t) and 7j;(x),
(9.4.1) can be reduced to the equation
x//(t)
+ q(t)f(x(t))
e(t).
(9.4.2)
We define h: JR ---+ JR by
h(x)
fox 7j;(u)du.
(9.4.3)
If 7j; is assumed to satisfy 7j;(x) > 0 for x of. 0, then clearly h

is increasing, continuously differentiable and xh( x) > 0 for x of. O.
Furthermore, the function g: h(JR) ---+ JR defined by
(9.4.4)
is continuous and satisfies xg(x) > 0 for x =I- O.
651
Theorem 9.4.1. Suppose l/J(X) > 0 for x i= O. If x = (t) is a

solution of (9.4.1) on some interval I, then z = h 0 ( t) is a solution of
the equation
(a(t)z'(t))' + q(t)g(z(t)) = e(t),
(9.4.5)
where hand g are as in (9.4.3) and (9.4.4).

Conversely, if z = ~(t) is a nontrivial solution of (9.4.5) on some
interval I, then x = h- 1 o~(t) is a nontrivial solution of equation (9.4.1)
on some interval J c I. If in additio111/)(0) i= 0, or z(t) i= 0 for all
tEl, then J = I.
Proof. It is easy to verify, by using (9.4.3) and (9.4.4), that z and x
defined above arc respectively solutions of equations (9.4.5) and (9.4.1) and
that z' = l/J(X)X' and J c I. In fact, this inequality together with (9.4.3)
shows that if 1jJ(0) = 0 and z vanishes at some tl E I, then x' may not
exist at t 1 ; in this case, J is a proper subset of I. The last statement
of the theorem also follows at once.
Corollary 9.4.1. Every oscillatory solution of (9.4.1) generates an oscillatory solution of equation (9.4.5).
Corollary 9.4.2. There is one-to-one correspondence between the nonoscillatory solutions of equations (9.4.1) and (9.4.5).
Corollary 9.4.3. Suppose e(t) = 0 and 1jJ(0) = O. If the solution
z(t) of (9.4.5) with z(to) = z'(t o) = 0 for every to?> 0 is unique, then
equation (9.4.1) has no nontrivial oscillatory solutions. If in addition (9.4.5)
is oscillatory, then equation (9.4.1) has no nontrivial continuable solutions.
Proof. Suppose x(t) is a nontrivial solution of (9.4.1), then by Theorem
9.4.1 and (9.4.3), z(t) = h(x(t)) is a nontrivial solution of equation (9.4.5)
such that
z'(t) = ljJ(X(t))x'(t).
(9.4.6)
By (9.4.3), x(t) vanishes if and only if z(t) vanishes. Since 1jJ(O) = 0

it follows from (9.4.6) that if Z(tl) = 0 for some tl ?> to ?> 0, then
z' (t 1 ) = 0 and hence from the uniqueness assumption z( t) == 0 and so is
x(t), which is a contradiction.
Example 9.4.1.
(9.4.7)
and let x(t) be its solution, then z(t) = x 3 (t)/3 is a solution of the linear
equation z"(t) + z(t) = O. Thus, z(t) = Asin(t + 0) for some constants
A and 0, and hence x(t) = Csin 1 / 3 (t + 0) where C = (3A)1/3. As
x'(t) does not exist for t = br - 0, k = 1,2"" equation (9.4.7) has no
nontrivial continuable solution.
Chapter 9
652
Example 9.4.2.

(9.4.8)
where
X 2n + 1
/(2n
is a positive integer and k is a constant.

and the associated equation is
+ 1)
z"(t)
+ (2n + 1)
C:)
z(t)
h(x)
Here,
(9.4.9)
0.
As equation (9.4.9) is oscillatory for k> 1/(8n+4) and nonoscillatory for

1/(8n + 4), by Corollary 9.4.3, equation (9.4.8) can have continuable
solutions only when k::: 1/(8n + 4), and hence by Corollary 9.4.2 no
nontrivial solution of (9.4.8) is oscillatory.
k:::
We now consider the unforced differential equation
(a(t)1j;(X(t))X'(t))'
for x
+ q(t)g(z(t))
0,
subject to the additional condition 1j;(X) >

tion h in (9.4.3) reduces (9.4.10) to
(a(t)z'(t))'
(9.4.10)
Ie 0.
The transforma-
+ q(t)f(x(t)) =
(9.4.11)
where g is defined in (9.4.4). If we now let

s
= R(t) -
it
to
du
a(u)
(9.4.12)
equation (9.4.11) is reduced to
y + R*(s)Q(s)g(y) =
where R*(s)
0,
(-
= djds)
(9.4.13)
= a[t(s)], Q(s) = q[t(s)J and y(s) = z[t(s)J.
We now state the following interesting results.
Theorem 9.4.2. Suppose 1j;(x) >

for x Ie 0, q(t) < on [tl, t2], tl 2::
and x(t) is a solution of equation (9.4.10) on [h, t2J such that X(tl) =
x'(tIl = 0. Then, x(t) = x'(t) = 0 for t E [tl,t2J if and only if
Jo+
where F(x) =
1j;(u) du
JF(u)
00
foX 1j;(u)f(u)du.
and
Jo-
1f;(u) du = _
JF(u)
00
653
Theorem 9.4.3. Suppose 'ljJ(x) > 0 for x -=J- 0, ft~o 'ljJ(u)du = oo,
and q(t) < 0 on [h, t 2 ], tl ;:::: O. Then equation (9.4.10) has a solution
x(t) such that limt-tT Ix(t)1 = 00 for some T E (tl, t2) if and only if
00
1j;(U)
d
'u <
Jl+F(u)
---r===~;:==:=
where F(x)
00
or
roo Jl1/;(+ F(u) d

'/1,)
io
,'/1,
> -
00,
= fox'ljJ(u)f(u)du.
Next we state two simple oscillation criteria for equation (9.4.10).

Theorem 9.4.4.
Suppose 1j;(X) > 0
conditions are satisfied
f'(x);:::: 0 for
for
x -=J- 0 and the following
x-=J-O,
(9.4.14)
00.
(9.4.15)

Theorem 9.4.5. Suppose 1j;(X) > 0 for x -=J- 0, conditions (9.4.14) and
(9.4.15) held, and f;IOO('ljJ(u)/f(u))du < 00. Then equation (9.4.10) is
oscillatory.
The next theorem describes the behavior of oscillatory solutions of equation (9.4.10) when 1/;(0) = O.
Theorem 9.4.6.
Suppose x(t) is a solution of equation (9.4.10) on
[tl' t2J such that x(t l ) = X(t2) = O. If 'ljJ(0) = 0 and q(t) does not
change sign on [tl: t2], then x(t) = 0 on [tl: t2J.
Proof. Suppose there exists t3 E (tl' t2) such that x( t 3) -=J- 0, then
there exist T I , T2 E [tl' t2J such that x(TIl = x(T2) = 0 and x(t) 1= 0
on (TI ,T2). Integrating equation (9.4.10) from TI to T 2, we obtain
T2
iT!
(a(s)'ljJ(x(s))x'(s))'ds
T2
iT!
q(s)f(x(s))ds = o.
As x(Td = x(T2) = 0 and 'ljJ(0) = 0, the first integral is zero, and hence
q(s)f(x(s))ds = o. As the integrand is of one sign and a continuous
function of t, it follows that q(t)f(x(t)) = 0 for all t E [TI ,T2], and
hence x(t) == 0 on [TI , T2J, which is a contradiction.
{i,'
Corollary 9.4.4. Suppose 'ljJ(0) = 0 and q(t) does not change sign.
Then the only oscillatory solution of equation (9.4.10) is the eventually
identically zero solution.
654
Chapter 9

1. Lemma 9.1.1 is taken from Leighton [21] and Swanson [26], Theorems
9.1.1 - 9.1.3 are due to Komkov [19]. Theorems 9.1.4 - !:U.8 are borrowed
from Graef and Spikes [13]. For related works we refer to Graef and Spikes
[12], Kreith [20] and Swanson [26,27].
2. Lemma 9.2.1 is due to Chanturija [7]. Theorems 9.2.1 - 9.2.4 arc
extracted from the work of Kawane et. a1. [17]. For more 011 this topic, we
refer to Kartsatos [15].
3. The results of Section 9.3 are taken from Graef [11]. For the related
subject we refer to Atkinson [1,2]' Bellman [3], Burlak [4], Burton et. a1.
[5,6], Hinton [14], Kauffman et. a1. [16], Knowles [18], Patula et. a.l. [23,24]'
Spikes [25], Weyl [28], and Wong et. a1. [29,30].
4. The results of Section 9.4 are due to Mahfoud and Rankin [22]. For
more general results we refer to Grace et. a1. [8-10].
9.6. References
1. F.V. Atkinson, Nonlinear extensions of limit point criteria, Math. Z.
130(1973), 297-312.
2. F.V. Atkinson, On second order differential inequalities, Proc. Roy. Soc.
Edinburgh Sec. A 72(1974), 109-127.
3. R. Bellman, Stability Theory of Differential Equations, McGraw Hill,
New York, 1953.
4. J. Burlak, On the nonexistence of Lz-solutions of nonlinear differential
equations, Proc. Edinburgh Math. Soc. 14(1965), 257-268.
5. T.A. Burton and R. Grimmer, On the asymptotic behavior of solutions
of x" + a(t)f(x) = e(t), Pacific J. Math. 41(1972), 43-55.
6. T.A. Burton and W.T. Patula, Limit circle results for second order
equations, Monatsh. Math. 81(1976), 185-194.
7. T.A. Chanturija, Some comparison theorems for higher order ordinary
differential equations, Bull. Acad. Polan. Sci. Ser. Sci. Math. Astronom.
Phys. 25(1977), 749-756 (in Russian).
9. S.R. Grace and B.S. Lalli, Integral averaging techniques for the oscillation of second order nonlinear differential equations, J. Math. Anal. Appl.
149(1990), 277-311.
SIAM J. Math. Anal. 15(1984), 1082-1093.
11. J.R. Graef, Limit circle criteria and related properties for nonlinear equations, J. Differential Equations 35(1980), 319-338.
655
12. J.R. Graef and P.W. Spikes, Sufficient conditions for nonoscillation of
a second order nonlinear differential equation, Proc. Amer. l\lath. Soc.
50(1975), 289-292.
13. J.R. Graef and P.W. Spikes, Comparison and nonoscillation results
for perturbed nonlinear differential equations, Ann. Mat. Pura Appl.
116(1978), 135-142.
14. D. Hinton, Limit point - Limit circle criteria for (py')' + qy = >"ky, in
Or'dina'ry and Partial Differential Equations, Lecture Notes in Math. 415,
Springer-Verlag, New York, 1974, 173-183.
15. A.G. Kartsatos, On the maintenance of oscillation of nth order equations
under the effect of small forcing term, J. Differential Equations 10(1971),
355-363.
16. R.M. Kauffman, T.T Read and A. Zettl, The Deficiency Index Problem for Powers of Ordinary Differential Expressions, Lecture Notes in
Math. 621, Springer-Verlag, New York, 1977.
17. N. Kawano, T. Kusano and M. Naito, Nonoscillatory solutions of
forced differential equations of the second order, J. Math. Anal. Appl.
90(1982), 323-342.
18. I. Knowles, On a limit-circle criterion for second order differential operators, Quart. J. Math. 24(1973), 451-455.
19. V. Komkov, A note on a generalization of the Sturm-Picone theorem,
Colloq. Math. XXXIX(1978), 173-176.
20. K. Kreith, Oscillation Theory, Lecture Notes in Math. 324, SpringerVerlag, New York, 1973.
21. W. Leighton, Comparison theorems for linear differential equations of
second order, Proc. Amer. Math. Soc. 13(1962), 603-610.
22. W.E. Mahfoud and S.M. Rankin, Some properties of solutions of
(r(t)'lI(x)x')' + a(t)f(x) = 0, SIAM J. Math. Anal. 10(1979), 49--54.
23. W.T. Patula and P. Waltman, Limit point classification of second order
linear differential equations, J. London Math. Soc. 8(1974), 209--216.
24. W.T. Patula and J.S.W. Wong, An LP analog of the Weyl alternative,
Math. Ann. 197(1972), 9-28.
25. P.W. Spikes, On the integrability of solutions of perturbed nonlinear
differential equations, Proc. Roy. Soc. Edinburgh Sec. A 77(1977), 309318.
27. C.A. Swanson, Picone's identity, Rend. Mat. 8(1975), 373-397.
28. H. Weyl, Uber gew6hnliche Differentialgleichungen mit Singularitaten
und die augeh6rige Entwicklung willkiirlicher Funktionen, Math. Ann.
68(1910), 220--269.
29. J.S.W. Wong, Remarks on the limit--circle classification of second order
differential operators, Quart. J. Math. 24(1973), 423-425.
30. J.S.W. Wong and A. Zettl, On the limit point classification of second
order differential equations, Math. Z. 132(1973), 297-304.
Chapter 10
Nonoscillation Theory
for Multivalued Differential
Equations
10.0. Introduction
In our previous chapters, we have presented several nonoscillation criteria for second order differential equations. In the present chapter, we shall
introduce nonoscillatory theory for second order differential and neutral inclusions. Our results rely on fixed point theorems for multivalued maps,
and on a compactness criterion.
10.1. Preliminaries
In this chapter, we shall provide nonoscillatory results for the second
order differential inclusions
(a(t)x'(t))' E e(t)
+ F(t, x(t)),
2: to 2: 0
(10.1.1)
and the neutral inclusions
:t (a(t)
:t
(x(t)
+ px[t -
T]))
F(t, x(t)),
2: to 2: o.
(10.1.2)
A nontrivial solution of (10.1.1), or (10.1.2) is called oscillatory if it has

arbitrarily large zeros, otherwise it is called nonoscillatory. To present our
nonoscillation criteria we shall need the following two fixed point theorems
for multivalued maps.
Theorem 10.1.1 (Ky-Fan's Fixed Point Theorem [4,10,11]). Let Q be
a nonempty, closed, convex subset of a Banach space E and F: Q -+
CK(Q) a upper semicontinuous, compact map; here CK(Q) denotes
Nonoscillation theory for multivalued differential equations
657
the family of nonempty convex compact subsets of Q. Then, there exists

x E Q with x E F(x).
Theorem 10.1.2 (Fit.zpatrick-Petryshyn Fixed Point Theorem [9]). Let
Q be a nonempty, closed, convex subset of a Banach space E and F : Q ---+
CK(Q) a upper semicontilluous, condensing map with F(Q) bounded.
Then, t.here exists :1: E Q with x E F(x).
We shall also need the followillg compactness criterion in B [T, (0) (the
Banach space of all continuous, bounded real valued functions 011 [T, (0)
endowed with the usual supremum norm, i.e., Ilull oo = snPtE[T,oo) lu(t)1
for u E B[T, (0).)
Theorem 10.1.3 [7]. Let E be all equicontinuous and uniformly bounded
subset of the Banach space B [T, (0). If E is equiconvergent at 00, then
it is relatively compact.
10.2. Differential Inclusions

In what follows with respect to the differential inclusion (1O.l.1), we
shall a.ssume that the functions a and e are single valued and F is a
multifunction. Furthermore, the followin~ conditions hold:
C([to, (0), lR+),

(ii) eEL 1 ([to, 00 ), lR+),
(iii) there exists an TJ E C([tD' (0), lR) such that (a(t) TJ'(t))' = e(t),
t 2: to,
(iv) F: [to, (0) x lR ---+ CK(lR) is a L1-Caratheodory multifunction: by
(i)
this we mean
(a). for each measurable u: [to, (0) ---+ lR the map t
measurable single valued selections,
(b). for a.e. t E [to, (0) the map u
F(t, u(t)) has
F(t, u) is upper semicontinuous,
(c). for each r > 0 there exists hr E L1 [to, (0) with IF(t, u)1 ::; hr(t)
for a.e. t E [to, (0) and all u E lR with lui::; r; here IF(t, u)1 =
sup{lzl: z E F(t, u)},
(v)
inftE[to,oo) TJ(t) >
-00.
Remark 10.2.1. In (iv), part (a) could be replaced by: the map t
F(t,u) is measurable for all u E JR..
Now, let (3 E lR be such that
(3 > -
inf
tE [to,oo)
TJ(t),
(10.2.1)
Chapter 10
658
and let d> 0 be such that
f3 + inf
'T)(t) 2 d.
tE[to,oo)
(10.2.2)
Theorem 10.2.1. Suppose (i) ~ (v) hold, and let f3 (respectively d) be

chosen as in (10.2.1) (respectively (10.2.2)). Also, assume that the following
three conditions are satisfied
F: [to, 00) x (0,00) -+ CK([O, 00)),
J 100
and
OO
M > f3 + sup 'T/(t)
M> d with
there exists
-1()
ass
(10.2.3)
(10.2.4)
t~to
sup
wE[d,Mj
IF(t,w)ldtds < 00.
(10.2.5)
Then there exists a nonoscillatory solution x of (10.1.1) for a.e.

with
lim (x(t) - 'T/(t)) =
t---+oo
f3 and
t 2 T
lim a(t)(x(t) - 'T/(t))' = 0;
t---+oo
here T is chosen as in (10.2.6).

Proof. From (10.2.5) there exists T 2 to such that
rOO _(1) 00
iT
ass
sup
wE[d,Mj
!F(t,w)ldtds::; M -
We wish to apply Theorem 10.1.1 with E
Q =
{xEB[T,oo):
[f3+ sup
'T/(t)].
t~T
= (B[T, 00), II . 1100)
d::;x(t)::;M
for
(10.2.6)
and
t2T}.
Clearly, Q is closed and convex. Also, if x E Q then for t 2 T, we have

from (10.2.3) and the definition of Q that 0::; u(t) ::; SUPwE[d,Mj!F(t, w)1
for each u(t) E F(t,x(t)). Define a mapping N: Q -+ P(E) (the power
set of E) by (here x E Q),
N x(t)
1
= f3 + 'T/(t) + itrOO a(s)
1
s
00
F(v, x( v))dvds
for t 2 T.
The Proposition 1.1 in [6, pp. 777] guarantees that N: Q -+ C(E); here
C(E) denotes the family of nonempty, convex subsets of E. We shall first
show that
(10.2.7)
N: Q -+ C(Q).
For notational purposes for any x E Q let
F(x)
= {UELl[T,oo): u(t)EF(t,x(t))
fora.e. tE[T,oo)}.
659
Let x E Q and take wEN x. Then there exists T E F(x) with
w(t) = f3 + ry(t)
1
a(s)
s
+ it
rOO
00
T(v)dvds
for t 2 T.
Then for t 2 T, we have from (10.2.6) that
w(t)
< f3 + sup ry(t) +

t?T
1 1
00
-1
00
a( s)
sup
wE [d,MJ
!FCt', w)ldvds
< f3 + sup ry(t) + [M - {f3 + sup ry(t)}] = M.

t?T
t?T
As a result w(t):S M for t 2 T for each wEN x. On the other hand

if t 2 T, we have
w(t) 2 f3
+ ry(t) 2
f3 + inf ry(t) 2 d.
t?to
As a result w(t) 2 d for t 2 T for each wEN x. Thus (10.2.7) holds.

Next, we shall show that
N :Q
C( Q)
(10.2.8)
is a compact map.
For this, we will use Theorem 10.1.3. For x E Q let G(x) = N x - ry.
Take any x E Q and wE G (x). Then there exists T E F(x) such that
w(t) = f3
+ it(=
a(s)
00
T(v)dvds
for t 2 T.
Now, since N: Q ~ C(Q), we have for t 2 T that w(t) < M

max {Id - f31, 1M - f31}, and hence for each wE G x it follows that
Ilwll=
sup
tE[T,=)
Iw(t)l:S M + max {Id - f3i, 1M - f31}
for each x E Q.
Thus, the set Y = {N x - (): x E Q} is a uniformly bounded subset of

B[T, (0). Also, for each t 2 T, we have
Iw(t)-f31 :S
1 1
00
1
-()
00
ass
sup
wE[d,MJ
IF(v,w)ldvds
(10.2.9)
for w E G x, x E Q. Now (10.2.5) and (10.2.9) guarantee that the set Y

is equiconvergent at 00. Also, for t}, t2 with T:S tl :S t2, we have
IW(t2) - w(tl)1 :S
t2
t,
00
1
sup !F(v,w)ldvds
a s s wE[d,MJ
- ()
Chapter 10
660
for tL' E G x, x E Q. Now Theorem 10.1.3 guarantees that Y is relatively

compact in B[T,oo), and as a result (10.2.8) holds.
It remains to show
N : Q -+ CK(Q)
is a upper semicontinuous map.
(10.2.10)
From (10.2.10) and [4, pp. 465] it is sufficient to show that the graph of
E giN) with (J.Ln, :r,,) -+ (J.L, :r);
here n E lNl = {1, 2, .}. We must show J.L E N x. Fix t E [T,oo).
Note that IlxilCXl S NI, Ilxnlloo S NI for n E lN 1 , since x, Xn E Q for
n E lN l . Also there exists Zn E F(xn) with
N, g(N), is closed. Consider (J.I n , xn)
J.Ln(t)
(3
tOO
1
a(s)
+ "l(t) + Jt
tOO
Js
zn(v)dvds
00 ill a(s)1 dsdv.

(3 + "l(t) + 1 Zn(V)
t
We must show that there exists
J.L(t) = (3 + "l(t)
(10.2.11)
F(x) with
00
1 (=
a(s)
u(v)dvds.
Js
Notice (iv) guarantees that there exists a hM ELI [T, (0) with IZn(s)1 S
hl\1(s) for a.e. s E [T,oo). Consider {zn}nElN, Take k E lNl and
k > t. From (10.2.11), we have
IJ.Ln(t) -
I lv
(3 - '(I(t) - .
I"
Zn(V)
a(s) dsdv
I S
Jk{= hM(V)
lv
t
a(s) dsdv.
(10.2.12)
Now, a standard result from the literature [14, Proposition 1.4] guarantees that Fk : B[T, k] = CrT, k] -+ Ll [T, k] is upper semicontinuous
with respect to the weak topology (w u.s.c.) and also weakly completely
continuous; here Fk is given by
Fk(W) = {u
Ll[T,k]: u(t) E F(t,w(t)) for a.e. t E [T,kJ}.
Since Zn E Fk (Xn) for n E IN 1, there exists a Uk E Ll [T, k] and a

subsequence of S of IN 1 with Zn converging weakly in Ll [T, k] to Uk
(i.e. Zn --' Uk in Ll [T, k]) as n -+ 00 in S. Now Xn -+ X in CrT, k]
and Zn --' Uk in Ll [T, k] as n -+ 00 in S, together with Zn E Fk (xn)
for n E Sand F k : CrT, k] -+ Ll [T, k] w-u.s.c., implies that
(10.2.13)
Note as well that IIxllk = SUPSE[T,k] Ix(s)1 S NI, Ilxnllk S NI for n E S,
and IUk(V)1 S hM(V) for a.e. v E [T, k]. Let n -+ 00 through S in

(10.2.12), to obtain
00
lv
- ()dsdv.
k
t a s
(10.2.14)
Similarly, we can show that there exists Uk+1 E Ll [T, k + 1] and a subsequence of S, say SI, with Zn --' Uk+l in Ll [T, k + 1] as n..--t 00 in
SI and with Uk+l E Fk+l (x). Of course this implies Zn --' Uk+l in
Ll[T,k] as n..--too in SI so Uk+l(V)=udv) fora.e. VE[T,k]. In
addition note that IUk+l(v)l:::; hM(V) for a.e. v E [T, k + 1]. Continue
and construct Uk+2, Uk+3,. For 1 E {k, k + I,} = 1Nk let uHv) be
any extension to [T,oo) of Uz with /ui(v)1 ~ hM(v) for a.e. v E (l,oo).
Also, let
~
Ft(w)
{z
L l [T,oo):
hM(V)
661
z(v) E F(v,w(v)) for a.e. v E [T,l],

Iz(v)1 ~ hM(V) for a.e. v E [T,oo)}.
Now, {uihEP is a weakly compact sequence in Ll[T, 00), (see [5 or 7]),

so there exists a subsequence which converges weakly to a function U E
Ll[T, 00). Note u(v) = Uk(V) for a.e. v E [T,k] since Uk+m(V) = Uk(V)
for a.e. v E [T, k]; here Tn E 1N l . This together with (10.2.14) yields
IIl(t) - /3 -1](t) -lk u(v)
a~) dSdVI ~
00
hM(V)
a(s) dsdv.
(10.2.15)
We next claim that
F (x)
n Ft
lEK
(x)
(here K
= {[T] + 1, [T] + 2, .})
(10.2.16)
(and :F (w) is nonempty, closed and convex). Note first that Ilxll oo :::; M
so IF(v, x(v))1 :::; hM(V) for a.e. v E [T,oo). Let Xk be the restriction
to the interval [T, k], k E 1N l , of x. Note [10,11 or 12] that
h (Xk) = {z
Ll[T,k]: z(v) E F(V,Xk(V)) for a.e. v E [T,k]}
is closed in L1[T, k] for all k E K. Let
:Fk (Xk) = {z
L1[T, 00) :
Z E:Fk (Xk) for v E [T,k]

and z(v) = 0 for v> k}.
It is immediate that :Fk (Xk) is a closed set in L1 [T, 00) for each k E K.
Let
Rk
{z E 1[T,oo):
z(v) = 0 for v E [T,k],

Iz(v)1 ~ hM(V) for a.e. v E (k, oo)}
Cbapter 10
662
and notice it is clear that :Fio (x) = :Fio (Xk) EEl Rk. It is clear that
:Fio (x) is a closed set in L1 [T, 00 ). Also, for each k E K, we have
:F (x) <;;; :Fio (x), and so :F (x) <;;; nCEK :Ft (x). On the other hand if
Z E :Fe' (x) for each IE K, then z(v) E F(v, x(v)) for a.e. v E [T, (0),
and so nCEK Ft (x) <;;; F (x). Thus, (10.2.16) holds and a.lso F (x) is a
closed subset of LI[T, (0). Now, since u belongs to nCEK Fc* (x) (note
for each I E K that u E Ft (x)), we have u E F (x). Let k -+ 00 in
(10.2.15), to obtain
J1(t)-/3-7](t)-
1
+1
t
and hence
J1(t) = /3 + 7](t)
00
u(v)
IV a(s)dsdv
1
00
a(s)
roo
is
u(l')dvds.
Consequently, Q(N) is closed, so N: Q -+ C(Q) is a upper semicontinuous map (see [4, pp. 465]); in fact (10.2.8) guarantees that N: Q -+
CK(Q).
Theorem 10.1.1 now guarantees that there exists x E Q with x E N x,
i.e., for every t 2: T, we have
x(t) E /3 + 7](t)
and so
(a(t)x'(t))'
In addition
<
Ix(t) - 7](t) - /31
e(t)
la(t)[x(t) - 7](t)l'1 -s;
00
roo
F(v, x(v))dvds,
+ F(t,x(t))
for a.e. t 2: T.
a(s)
1 1
00
-1
a(s)
-+0
and
is
00
sup
wE[d,MJ
00
sup
wE[d,MJ
IF(v, w)ldvds
ast-+oo
IF(v, w)ldv -+
as
t -+
00.
We shall next show that an existence result for (10.1.1) can be obtained
if (10.2.3) is replaced by
F: [to, (0) x (0,00) -+ CK((-oo, 0]).
(10.2.17)
Theorem 10.2.2.
Suppose (i) - (v) and (10.2.17) hold, and let /3
(respectively d) be chosen as in (10.2.1) (respectively (10.2.2)). Also,
assume that the following two conditions are satisfied
there exist
K >1
and
d
M > K
with
M 2: /3 + sup 7](t) (10.2.18)

t;:.:to
663
and
OG
o.(s)
(XJ
W(t, w)ldtd8 < 00.
sup
Js
wEld/K,MI
(10.2.19)
Then there exists a nonoscillatory solution x of (10.1.1) for a.e. t::;> T

with limt-+oo (x(t) -ry(t)) = (3 and limt-+oo a(t)(x(t) -ry(t))' = 0; here
T is chosen as in (10.2.20).
Proof. From (10.2.19) there exists T 2.'- to with
oo
0.(8)
[00
Js
W(t,w)ldtd8::;
sup
We will apply Theorem 10.1.1 with E
{xEB[T,oo):
(K;;
wEld/K,MI
=
(E[T, 00),
1) d.
II . 1100)
(10.2.20)
and
dJK::;x(t)::;M for t::;>T}.
Let Nand F be as in Theorem 10.2.1. First, we shall show that
N: Q
-7
(10.2.21 )
C(Q).
Let x E Q and take wEN x. Then there exists T E F(x) such that
1 1
J= 1=
w(t) = (3 + ry(t) +
If t::;> T,
w(t)
we also have
>
(3
+ ry(t) -
00
00
ass
-1
a(s)
> (3+ inf 7/(t)t?:.T
-1(
.)
for
sup
W(t, w)ldtd8
::;> d-
(K-1)
- - d = Kd
WEld/K,MI
- d
(-K-1)
K
T(1,)dl,d8
As a result w(t)::;> dJK for t::;> T for each wEN x. On the other
hand for t 2.'- T, we have w(t)::; (3 + ry(t) ::; (3 + SUPt>T ry(t)::; M. Thus,
w(t) ::; M for t ::;> T for each '11' E N x. Hence, (10.2.21) holds. Essentially
the same reasoning as in Theorem 10.2.1 guarantees that N: Q -7 CK(Q)
is a upper semicontinuous, compact map. Now, Theorem 10.1.1 proves the
result.
In our next result, we will show that the assumption (10.2.17) (respectively, (10.2.3)) in Theorem 10.2.2 (respectively, Theorem 10.2.1) can be
removed if we combine the analysis of both the theorems. In this case
(10.2.18) has to be adjusted slightly.
Theorem 10.2.3. Suppose (i) ~ (v) hold, and let (3 (respectively d)
be chosen as in (10.2.1) (respectively, (10.2.2)). Also, assume that the
following two conditions are satisfied
there exist
K > 1 and
M>K
with
M > (3 + sup ry(t) (10.2.22)

t?:.to
664
Chapter 10
and
oo
a( s)
/00
S
sup
wE Id/ K,Mj
IF(t, 'W)ldtds < 00.
(10.2.23)
Then there exists a nonoscillatory solution x of (10.1.1) for a.e. t 2: T

with limHoo (;c(t) ~ Il(t)) = (3 and limHoo a(t)(x(t) ~ 71(t))' = 0; here
T is chosen as in (10.2.24).
Proof.
roo
iT
From (10.2.23) there exists T 2: to such that

1
00
a(s)
sup
WEld/ K,Mj
IF(t, 'W)ldtds
(10.2.24)
Let Q and N be as in Theorem 10.2.2. Let x E Q and take 'W EN x.

As in Theorem 10.2.2, we find w(t) 2: d/K for t 2: T and as in Theorem
10.2.1, we have 'W(t):<:: M for t 2: T. Thus, N: Q ---+ C(Q).
Remark 10.2.2. Minor adjustments in the analysis of this section would

enable us to discuss the more general differential inclusion
(a(t)(b(t)x(t))')'
e(t)
+ F(t, x(t)),
t 2: to 2: o.
10.3. Neutral Inclusions

In this section, we shall discuss the neutral inclusion (10.1.2), where
the function a is single valued, p and T are constants, and F
is a multifunction. In what follows, we shall assume that the following
conditions hold:
(vi) T E IR+,
(vii) a E C([to,oo),JR+),
(viii) F: [to, 00) x lR ---+ CK(JR) is a LI-CaratModory multifunction.
Theorem 10.3.1.
Suppose (vi) - (viii) hold.
following three conditions are satisfied
F:[to,oo) x (0,00) -,
Ipl =I
and
there exists K
> 0 such that
CK((~oo,O])
1
~
a(s)
00
(10.3.1)
(10.3.2)
J 1
oo
Also, assume that the
sup
wE[K/2,Kj
IF(t, 'W)ldtds < 00.

(10.3.3)
665
Nonoscillation theory for multi valued differential equations
Then there exists a bounded nonoscillatory solution x of (10.1.2) for a.e.

t ;:: T ;:: to with T suitably chosen (see the proof of the theorem).
Remark 10.3.1 One cannot expect an analog of Theorem 10.3.1 for the
case Ipi = 1 even in the neutral equation case [8, Chapter 5] (see also [2,
Chapter 3] for some partial results when Ipi = 1).
Proof. We need to consider the two cases
Ipi <
Case (I).
(= 1
JT
Let TI
I . 11(0)
a(s)
=T -
Ipi <
1. Choose T 2: to so thFlt t -
1=
Ipl >
1 separately.
> to for t 2: T and
IF(t,w)ldtds -s: 4'(l-lpl)K.
( 10.3.4)
We wish to apply Theorem 1O.l.2 with E
= (B[TI' (0),
T.
1 and
sup
wE[K/2,K]
and
Define the single valued map NI : Q --+ E

N2 : Q --+ P(E) as follows (here x E Q):
_
-
4'(1
and
N2 x(t)
4'(1
+ p)K -
px[T - T],
+ p)K -
px[t - T],
r 1=
and the multivalued map
{0 } ,
TI
- JT
a(s)
-s: t -s: T
F(v,x(v))dl'ds,
t 2: T.
First, we shall show that

(10.3.5)
For notational purposes for any x E Q let
F(x)
{u
LIm, (0): u(t)
F(t, x(t)) for a.e. t
E [TI' oo)}.
Let x E Q so that K/2 -s: x(t) -s: K for t E [TI' (0), and take wE N2 x.
Then there exists TO E F( x) with
w(t) =
r a(s) 1= To(v)dV ds ,
{O},
TI
- JT
-s: t -s: T
s
t;:: T
666
Chapter 10
Now, we need to consider two subcases, namely, O:s; p < 1 and

P
- 1<
< O.
Subcase (1).
Nl x(t)
O:s; P < 1. If Tl :s; t :s; T, then clearly

3
+ wet)
4(1
+ p)K -
px[T - T]
~ ~(l+P)K-PK
(~-~P)K
>
and
If t ~ T, then (10.3.1) implies

3
3
K
Nl x(t)+w(t) ~ 4(1+p)K- p x[t-T] ~ 4(1+p)K-pK > 2'
and (10.3.4) implies

Nl x(t)
+ wet)
+ p)K -
+ 4(1 -
:s;
4(1
<
3
K
1
4(1+p)K- P2+4(1-p)K
px[t - T]
As a result K/2:S; Nl x(t) + 'l1J(t) :s; K for t

Thus, (10.3.5) holds in this case.
p)K
= K.
Tl for each w E N2 x.
Sub case (2). -1 
+ p)K -
px[T - T]
~(1 + p)K _ pK
4
(~+ ~p) K >
and
If t ~ T,
Nl x(t)
then (10.3.1) implies
wet)
~ ~(1 + p)K -
px[t - T]
(~+ ~p) K ~ ~,
>
and (10.3.4) implies

Nl x(t)
+ wet) < 4 (1 + p)K -
pK +
4(1 + p)K
K.
667
As a result K/2:S; Nl x(t) + w(t) :s; K for t ~ Tl for each wE N2 x.

Thus, (10.3.5) holds in this case also.
Essentially the same reasoning as in Theorem 10.2.1 guarantees that
N2 : Q -+ C(E)
is a upper semicontinuous, compact map.
(10.3.6)
Next, we claim that

N1
To see this notice for
whereas if t
Q -+ E
Xl,
X2
is a contractive map.
(10.3.7)
Q and Tl :s; t :s; T, we have
T, we have
so in view of ipi < 1, (10.3.7) holds.

Now, (10.3.5), (10.3.6) and (10.3.7) imply that
N: Q -+ CK(Q)
is a upper semicontinuous, condensing map. (10.3.8)
Thus, Theorem 10.1.2 implies that there exists

N2 x. Hence, for t ~ T, we have
3
x(t) E -(1
4
+ p)K -
px[t - T] -
it 1
T
1
-()
ass
00
Q with x E Nl
F(v, x(v))dvds.
This completes the proof of Case (I).

Case (II). ipi
roo
iT
> 1. Choose T
1
a(s)
00
~ to so that t -
sup
wE[K/2,KJ
> to for t
~ T
and
iF(t, w)idtds :s; -4(ipi-1)K.
Let Tl = T - T, and E and Q be as in Case (I). Define Nl : Q -+ E

and N 2 : Q -+ P(E) as follows (here x E Q):
3(1 + p)
{4
-p-
Nl x(t) =
K - px[T + T],
P) K - -X[t+T],
1
-3(1+
-4
668
Chapter 10
and
N2 x(t) =
{O},
1
-~
Tl '5:. t '5:. T
jt+r T+T
o.(s)
00
F(v, x(vdvds,
T.
Now, a slight modification of the argument in Case (I) guarantees that

N : Q -t CK(Q) is a upper semicontinuous, condensing map. Thus,
Theorem 10.1.2 is applicable. This completes the proof.

1. All the results of this chapter are based on the recent work of Agarwal,
Grace and O'Regan [1]. These results improve the corresponding ones for
the single valued case presented in earlier chapters and in [2]. Further
applications of the fixed point theorems used in this chapter are available
in [3,13].
10.5. References
1. R.P. Agarwal, S.R. Grace and D. O'Regan, On nonoscillatory solutions of differential inclusions, Proc. Amer. Math. Soc., to appear.
2. R.P.Agarwal, S.R. Grace and D. O'Regan, Oscillation Theory for
3. R.P. Agarwal, M. Meehan and D. O'Regan, Fixed Point Theory and
Applications, Cambridge University Press, Cambridge, 2001.
4. C.D. Aliprantis and K.C. Border, Infinite Dimensional Analysis, Springer, Berlin, 1994.
5. J.P. Aubin and A. Cellina, Differential Inclusions, Springer, Berlin,
1984.
5. M. Cecchi, M. Marini and P. Zecca, Existence of bounded solutions
for multivalued differential systems, Nonlinear Analysis 9(1985), 775~785.
7. N. Dunford and J. Schwartz, Linear Operators, Interscience, New York,
1958.
8. L.H. Erbe, Q.K. Kong and B.G. Zhang, Oscillation Theory for Functional Differential Equations, Marcel Dekker, New York, 1995.
9. P.M. Fitzpatrick and W.V. Petryshyn, Fixed point theorems for multivalued noncompact acyclic mappings, Pacific Jour. Math. 54(1974),
17~23.
10. M. Frigon, Theoremes d'existence de solutions d'inclusions differentielles,

in Topological Methods in Differential Equations and Inclusions (edited by
A. Granas and M. Frigon), NATO AS! Series C, Vol 472, Kluwer, Dordrecht, 1995, 51~87.
669
11. A. Lasota and Z. Opial, An application of the Kututani-Ky Fan theorem

in the theory of ordinary differential equations, Bull. Acad. Polan. Sci.
Ser. Sci. Math. Astron. Phys. 13(1965), 781-786.
12. D. O'Regan, Integral inclusions of upper semi-continuous or lower semi-
continuous type, Proc. Amer. Math. Soc. 124(1996), 2391-2399.
13. D. O'Regan and R. Precup, Theorems of Leray-Schauder Type and
Applications, Gordon & Breach, Amsterdam, 2001.
14. T. Pruszko, Topological degree methods in multivalued boundary value
problems, Nonlinear Analysis 5(1981), 953-973.
Subject Index
Arzela-Ascoli theorem
asymptotic behavior
averaging technique
Banach lattice
Besicovitch almost periodic functions
Besicovitch seminorm
bounded solution
compactness criterion
comparison theorems
conjugate
continuability
converges to zero
damped equation
decaying nonoscillatory solution
differential inclusion
differential system
differential system with forcing term
disconjugate
distance of zeros
Emden-Fowler equation
Euler differential equation
Euler-Lagrange equation
Fitzpatrick-Petryshyn fixed point theorem
forced equation
functional differential equations
functionally commutative
general damped equation
generalized Emden-Fowler equation
generalized Euler equation
generalized Prufer transform
generalized Riccati equation
generalized sine function
Gronwall inequality
half-linear differential equation
11
549,559,562,567,589,608
40
540
107
107
4,5,6,8,10,562
11,657
15
469
2
6,7,8,10
205,376
262
656,657
503
549
467,504
210
36,433,457,493,641
17,23,38,69
94
657
82,205,632
218,237
533
274
559
467
164
209,467,470
164
2
94,444,466,552
Subject Index
Hartman's oscillation criterion
Hille theorem
Hille-Wintner comparison theorem
integrably small
integral averaging technique
interval criteria
interval oscillation
Kummer transformation
Ky-Fan fixed point theorem
Lebesgue dominated convergence theorem
left dis focal
Leighton's oscillation criterion
Levin-type comparison theorem
Liapunov function
Liapunov inequality
Liapunov second method
limit circle type
limit point type
linearization result
maximal solution
minimal solution
mixed type equation
neutral equation
neutral inclusions
nonlinear damping term
nonoscillatory
nonoscillatory characterization
nonprincipal solution
Olech-Opial-Wazewski method
operator-valued differential equation
oscillatory
Picone identity
positive decaying solution
positive increasing solution
positive solution
prepared solution
principal solution
proper solution
quickly oscillatory
Riccati differential equation
Riccati inequality
671
90
601
332,540
70
168
179
64,66
90
656
11
213
90
100
477
105
476
639
639
339,429
342,636,637
342,636,637
257
233
656,664
362
1,492
15
467
457
540
1,492,504
14
595
608
585,614
504
467
113,585
107
501
506
672
Riccati integral equation
Riccati transformation
Riccati variable
right disfoca.l
right-hand derivative
Schauder fixed point theorem
Schaudcr-Tychonov fixed point theorem
self-conjugate
singular Emden-Fowler equation
singular Emden-Fowler system
solution
strongly sublinear
strongly superlinear
Sturm theorem
Sturmian identity
Sturm-Picone theorem
sublinear
superlinear
superlinear forced equation
two-dimensional differential system
unforced equation
variational principle
weighted function
weighted-average oscillation criteria
weighted-integral condition
Willett's oscillation criterion
Wirtinger type inequality
Wronskian
subject Index
22,501,505
466
544
213
477
11
11
504
584,602,606
614,620
1,4
448
274
13,94,99,626
14
15,625,628
376,377,433,498,569
274,296,493,566
358
492
82
467
46
188
45
90
96
14

Oscillation Theory For Second Order Linear, Half-Linear, Superlinear and Sublinear Dynamic Equations PDF

Uploaded by

Document Information

Original Title

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

Oscillation Theory For Second Order Linear, Half-Linear, Superlinear and Sublinear Dynamic Equations PDF

Uploaded by

Copyright:

Available Formats

Oscillation Theory for Second Order

Linear, Half-Linear, Superlinear and Sublinear Dynamic

Oscillation Theory for Second

Donal 0' Regan

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

Printed on acid-free paper

AII Rights Reserved

Oscillation and N onoscillation of Linear

Oscillation and N onoscillation of

Oscillation Theory for Superlinear

Existence of Nonoscillatory Solutions and Comparison Results

Superlinear Forced Oscillations

Oscillation Theory for Sublinear

Further Results on the Oscillation of

Some Definitions and Notation

Oscillation Results for

Asymptotic Behavior of Solutions of

8.2.1. Positive Solutions of Equation (8.2.1)

8.2.3. Positive Solutions of Singular Emden-Fowler-Type Systems

Chapter 10 N onoscillation Theory for Multivalued

F(t, x(t), x'(t), xl/(t)) = 0,

1.1. Continuability, Boundedness

+ h(t,x(t),x'(t + q(t)f(x(tg(x'(t = e(t,x(t),x'(t,

= att) [-a'(t)y - h(t,x,y) - q(t)f(x)g(y) + e(t,x,y)].

Let p+(t) = max{p(t) , O} and p- (t) = maxi -p~t), O} so that p(t) =

and there are nonnegative constants m and n such that

be an interval of real numbers, and

where c is a nonnegative constant, and u, q E C(I, lR+). Then,

V(x, y, t) = q(t) G(y)

+ a(t) [F(x) + KJ,

[q;~~!~- G(y) + a(:~2(t) ;~)

+ m a(t)q(t) + n a(t)q(t) G(y).

for some constant Kl > O. Now, from Gronwall's inequality, we find

[[q'(s)]- +nr(s)] dS)

where K2 is a constant. Thus, G(y(t)) is bounded on [to, T), and

1. If e(t, x, y) == 0 in Theorem 1.1.1, then condition (1.1.3) can be dropped.

Now, in what follows by the term 'solution' we shall mean a continuously

Theorem 1.1.2. Suppose (1.1.4) and (1.1.5) hold with n> 0,

a'(t) 2: 0 and a(t)

then all solutions of equation (1.1.1) are bounded.

~~!~ [F(x) + K] + G(y),

q'(t) (q(t) [F(x)

+ K] a(t) + a(t) e(t,x,y) g(y)

Integrating the above inequality from to to t and using (1.1.5), we find

~ + n q(t) [F(x(t)) + K] <

Now, applying Gronwall's inequality, we get

where Kl > 0 is a constant. Hence, nq(t)F(x(t))/a(t):s: K1q(t)/q(to),

Proof. Condition (1.1.6) implies that there exists a constant A >

By condition (1.1.8), F(x) 2': -K for some constant K> 0. If M 2': 1,

!~!~ [F(x) + K] + G(y) + A + B.

Then since M 2': 1, we have

= q'(t) (q(t) [F(x) + K] + ~~)

+ [a'(t)]- (q(t) [F(x)

~g; (~~!~[F(X)+K]+ ~(A+B)+G(Y))

(q'(t) +M[a'(t)]-) (q(t)[F(X)+K]+A+B+G(Y)).

Integrating V'(t) from to to t, we get

< V (to) exp (

Then since M < 1, we have

~~j (~~!~[F(X)+K]+ ~(A+B)+G(Y))

The remainder of the proof follows as before.

V(t) ::; V(to) +

-(-) V(s)ds + -In -(-) .

Hence, it follows that

j (!~[F(X)+K]+ ~(A+B)+G(Y))

p(t) [q(t) + ka(t)h2(t) - (a(t)h(t' - (p(t)h*(t]